Model Deployment : Classifying Brain Tumors from Magnetic Resonance Images by Leveraging Convolutional Neural Network-Based Multilevel Feature Extraction and Hierarchical Representation¶
- 1. Table of Contents
- 1.1 Data Background
- 1.2 Data Description
- 1.3 Data Quality Assessment
- 1.4 Data Preprocessing
- 1.5 Data Exploration
- 1.6 Predictive Model Development
- 1.6.1 Pre-Modelling Data Preparation
- 1.6.2 Convolutional Neural Network Sequential Layer Development
- 1.6.3 CNN With No Regularization Model Fitting | Hyperparameter Tuning | Validation
- 1.6.4 CNN With Dropout Regularization Model Fitting | Hyperparameter Tuning | Validation
- 1.6.5 CNN With Batch Normalization Regularization Model Fitting | Hyperparameter Tuning | Validation
- 1.6.6 CNN With Dropout and Batch Normalization Regularization Model Fitting | Hyperparameter Tuning | Validation
- 1.6.7 Model Selection
- 1.6.8 Model Testing
- 1.6.9 Model Inference
- 1.7 Predictive Model Deployment Using Streamlit and Streamlit Community Cloud
- 2. Summary
- 3. References
1. Table of Contents ¶
1.1 Data Background ¶
1.2 Data Description ¶
In [1]:
##################################
# Loading Python Libraries
##################################
##################################
# Data Loading, Data Preprocessing
# and Exploratory Data Analysis
##################################
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import matplotlib.cm as cm
from matplotlib.offsetbox import OffsetImage, AnnotationBbox
%matplotlib inline
import tensorflow as tf
import keras
from PIL import Image
from glob import glob
import cv2
import os
import random
import math
##################################
# Model Development
##################################
from keras import backend as K
from keras import regularizers
from keras.models import Sequential, Model,load_model
from keras.layers import Input, Activation, Dense, Dropout, Flatten, Conv2D, MaxPooling2D, MaxPool2D, AveragePooling2D, GlobalMaxPooling2D, BatchNormalization
from keras.optimizers import Adam, SGD
from keras.callbacks import ReduceLROnPlateau, EarlyStopping, ModelCheckpoint
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.utils import img_to_array, array_to_img, load_img
from math import ceil
##################################
# Model Evaluation
##################################
from keras.metrics import PrecisionAtRecall, Recall
from sklearn.metrics import confusion_matrix
from sklearn.metrics import precision_recall_fscore_support, accuracy_score
In [2]:
##################################
# Setting random seed options
# for the analysis
##################################
def set_seed(seed=123):
np.random.seed(seed)
tf.random.set_seed(seed)
keras.utils.set_random_seed(seed)
random.seed(seed)
tf.config.experimental.enable_op_determinism()
os.environ['TF_DETERMINISTIC_OPS'] = "1"
os.environ['TF_CUDNN_DETERMINISM'] = "1"
os.environ['PYTHONHASHSEED'] = str(seed)
set_seed()
In [3]:
##################################
# Filtering out unncessary warnings
##################################
import warnings
warnings.filterwarnings('ignore')
In [4]:
##################################
# Defining file paths
##################################
DATASETS_ORIGINAL_PATH = r"datasets\Brain_Tumor_MRI_Dataset"
DATASETS_FINAL_TRAIN_PATH = r"datasets\Brain_Tumor_MRI_Dataset\Training"
DATASETS_FINAL_TEST_PATH = r"datasets\Brain_Tumor_MRI_Dataset\Testing"
MODELS_PATH = r"models"
PARAMETERS_PATH = r"parameters"
PIPELINES_PATH = r"pipelines"
In [5]:
##################################
# Defining the image category levels
# for the training data
##################################
diagnosis_code_dictionary_train = {'Tr-no': 0,
'Tr-noTr': 0,
'Tr-gl': 1,
'Tr-glTr': 1,
'Tr-me': 2,
'Tr-meTr': 2,
'Tr-pi': 3,
'Tr-piTr': 3}
##################################
# Defining the image category descriptions
# for the training data
##################################
diagnosis_description_dictionary_train = {'Tr-no': 'No Tumor',
'Tr-noTr': 'No Tumor',
'Tr-gl': 'Glioma',
'Tr-glTr': 'Glioma',
'Tr-me': 'Meningioma',
'Tr-meTr': 'Meningioma',
'Tr-pi': 'Pituitary',
'Tr-piTr': 'Pituitary'}
In [6]:
##################################
# Consolidating the image path
# for the training data
##################################
imageid_path_dictionary_train = {os.path.splitext(os.path.basename(x))[0]: x for x in glob(os.path.join("..", DATASETS_FINAL_TRAIN_PATH, '*','*.jpg'))}
In [7]:
##################################
# Taking a snapshot of the dictionary
# for the training data
##################################
dict(list(imageid_path_dictionary_train.items())[0:5])
Out[7]:
{'Tr-glTr_0000': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Training\\glioma\\Tr-glTr_0000.jpg',
'Tr-glTr_0001': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Training\\glioma\\Tr-glTr_0001.jpg',
'Tr-glTr_0002': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Training\\glioma\\Tr-glTr_0002.jpg',
'Tr-glTr_0003': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Training\\glioma\\Tr-glTr_0003.jpg',
'Tr-glTr_0004': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Training\\glioma\\Tr-glTr_0004.jpg'}
In [8]:
##################################
# Consolidating the information
# from the training datas
# into a dataframe
##################################
mri_images_train = pd.DataFrame.from_dict(imageid_path_dictionary_train, orient = 'index').reset_index()
mri_images_train.columns = ['Image_ID','Path']
classes = mri_images_train.Image_ID.str.split('_').str[0]
mri_images_train['Diagnosis'] = classes
mri_images_train['Target'] = mri_images_train['Diagnosis'].map(diagnosis_code_dictionary_train.get)
mri_images_train['Class'] = mri_images_train['Diagnosis'].map(diagnosis_description_dictionary_train.get)
In [9]:
##################################
# Performing a general exploration of the training data
##################################
print('Dataset Dimensions: ')
display(mri_images_train.shape)
Dataset Dimensions:
(5712, 5)
In [10]:
##################################
# Listing the column names and data types
# for the training data
##################################
print('Column Names and Data Types:')
display(mri_images_train.dtypes)
Column Names and Data Types:
Image_ID object Path object Diagnosis object Target int64 Class object dtype: object
In [11]:
##################################
# Taking a snapshot of the training data
##################################
mri_images_train.head()
Out[11]:
| Image_ID | Path | Diagnosis | Target | Class | |
|---|---|---|---|---|---|
| 0 | Tr-glTr_0000 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma |
| 1 | Tr-glTr_0001 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma |
| 2 | Tr-glTr_0002 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma |
| 3 | Tr-glTr_0003 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma |
| 4 | Tr-glTr_0004 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma |
In [12]:
##################################
# Performing a general exploration of the numeric variables
# for the training data
##################################
print('Numeric Variable Summary:')
display(mri_images_train.describe(include='number').transpose())
Numeric Variable Summary:
| count | mean | std | min | 25% | 50% | 75% | max | |
|---|---|---|---|---|---|---|---|---|
| Target | 5712.0 | 1.465336 | 1.147892 | 0.0 | 0.0 | 1.0 | 3.0 | 3.0 |
In [13]:
##################################
# Performing a general exploration of the object variables
# for the training data
##################################
print('Object Variable Summary:')
display(mri_images_train.describe(include='object').transpose())
Object Variable Summary:
| count | unique | top | freq | |
|---|---|---|---|---|
| Image_ID | 5712 | 5712 | Tr-pi_1440 | 1 |
| Path | 5712 | 5712 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\p... | 1 |
| Diagnosis | 5712 | 8 | Tr-no | 1585 |
| Class | 5712 | 4 | No Tumor | 1595 |
In [14]:
##################################
# Performing a general exploration of the target variable
# for the training data
##################################
mri_images_train.Class.value_counts()
Out[14]:
Class No Tumor 1595 Pituitary 1457 Meningioma 1339 Glioma 1321 Name: count, dtype: int64
In [15]:
##################################
# Performing a general exploration of the target variable
# for the training data
##################################
mri_images_train.Class.value_counts(normalize=True)
Out[15]:
Class No Tumor 0.279237 Pituitary 0.255077 Meningioma 0.234419 Glioma 0.231268 Name: proportion, dtype: float64
In [16]:
##################################
# Defining the image category levels
# for the training data
##################################
diagnosis_code_dictionary_test = {'Te-no': 0,
'Te-noTr': 0,
'Te-gl': 1,
'Te-glTr': 1,
'Tr-me': 2,
'Te-meTr': 2,
'Te-pi': 3,
'Te-piTr': 3}
##################################
# Defining the image category descriptions
# for the training data
##################################
diagnosis_description_dictionary_test = {'Te-no': 'No Tumor',
'Te-noTr': 'No Tumor',
'Te-gl': 'Glioma',
'Te-glTr': 'Glioma',
'Te-me': 'Meningioma',
'Te-meTr': 'Meningioma',
'Te-pi': 'Pituitary',
'Te-piTr': 'Pituitary'}
In [17]:
##################################
# Consolidating the image path
# for the testing data
##################################
imageid_path_dictionary_test = {os.path.splitext(os.path.basename(x))[0]: x for x in glob(os.path.join("..", DATASETS_FINAL_TEST_PATH, '*','*.jpg'))}
In [18]:
##################################
# Taking a snapshot of the dictionary
# for the testing data
##################################
dict(list(imageid_path_dictionary_test.items())[0:5])
Out[18]:
{'Te-glTr_0000': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Testing\\glioma\\Te-glTr_0000.jpg',
'Te-glTr_0001': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Testing\\glioma\\Te-glTr_0001.jpg',
'Te-glTr_0002': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Testing\\glioma\\Te-glTr_0002.jpg',
'Te-glTr_0003': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Testing\\glioma\\Te-glTr_0003.jpg',
'Te-glTr_0004': '..\\datasets\\Brain_Tumor_MRI_Dataset\\Testing\\glioma\\Te-glTr_0004.jpg'}
In [19]:
##################################
# Consolidating the information
# from the testing datas
# into a dataframe
##################################
mri_images_test = pd.DataFrame.from_dict(imageid_path_dictionary_test, orient = 'index').reset_index()
mri_images_test.columns = ['Image_ID','Path']
classes = mri_images_test.Image_ID.str.split('_').str[0]
mri_images_test['Diagnosis'] = classes
mri_images_test['Target'] = mri_images_test['Diagnosis'].map(diagnosis_code_dictionary_test.get)
mri_images_test['Class'] = mri_images_test['Diagnosis'].map(diagnosis_description_dictionary_test.get)
In [20]:
##################################
# Performing a general exploration of the testing data
##################################
print('Dataset Dimensions: ')
display(mri_images_test.shape)
Dataset Dimensions:
(1311, 5)
In [21]:
##################################
# Listing the column names and data types
# for the testing data
##################################
print('Column Names and Data Types:')
display(mri_images_test.dtypes)
Column Names and Data Types:
Image_ID object Path object Diagnosis object Target float64 Class object dtype: object
In [22]:
##################################
# Taking a snapshot of the testing data
##################################
mri_images_test.head()
Out[22]:
| Image_ID | Path | Diagnosis | Target | Class | |
|---|---|---|---|---|---|
| 0 | Te-glTr_0000 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma |
| 1 | Te-glTr_0001 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma |
| 2 | Te-glTr_0002 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma |
| 3 | Te-glTr_0003 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma |
| 4 | Te-glTr_0004 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma |
In [23]:
##################################
# Performing a general exploration of the numeric variables
# for the testing data
##################################
print('Numeric Variable Summary:')
display(mri_images_test.describe(include='number').transpose())
Numeric Variable Summary:
| count | mean | std | min | 25% | 50% | 75% | max | |
|---|---|---|---|---|---|---|---|---|
| Target | 1015.0 | 1.20197 | 1.245741 | 0.0 | 0.0 | 1.0 | 3.0 | 3.0 |
In [24]:
##################################
# Performing a general exploration of the object variables
# for the testing data
##################################
print('Object Variable Summary:')
display(mri_images_test.describe(include='object').transpose())
Object Variable Summary:
| count | unique | top | freq | |
|---|---|---|---|---|
| Image_ID | 1311 | 1311 | Te-pi_0299 | 1 |
| Path | 1311 | 1311 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\pi... | 1 |
| Diagnosis | 1311 | 8 | Te-no | 395 |
| Class | 1311 | 4 | No Tumor | 405 |
In [25]:
##################################
# Performing a general exploration of the target variable
# for the testing data
##################################
mri_images_test.Class.value_counts()
Out[25]:
Class No Tumor 405 Meningioma 306 Glioma 300 Pituitary 300 Name: count, dtype: int64
In [26]:
##################################
# Performing a general exploration of the target variable
# for the testing data
##################################
mri_images_test.Class.value_counts(normalize=True)
Out[26]:
Class No Tumor 0.308924 Meningioma 0.233410 Glioma 0.228833 Pituitary 0.228833 Name: proportion, dtype: float64
1.3 Data Quality Assessment ¶
In [27]:
##################################
# Counting the number of duplicated images
# for the training data
##################################
mri_images_train.duplicated().sum()
Out[27]:
np.int64(0)
In [28]:
##################################
# Gathering the number of null images
##################################
mri_images_train.isnull().sum()
Out[28]:
Image_ID 0 Path 0 Diagnosis 0 Target 0 Class 0 dtype: int64
In [29]:
##################################
# Counting the number of duplicated images
# for the testing data
##################################
mri_images_test.duplicated().sum()
Out[29]:
np.int64(0)
In [30]:
##################################
# Gathering the number of null images
##################################
mri_images_test.isnull().sum()
Out[30]:
Image_ID 0 Path 0 Diagnosis 0 Target 296 Class 0 dtype: int64
1.4 Data Preprocessing ¶
In [31]:
##################################
# Including the pixel information
# of the actual images in array format
# for the training data
# into a dataframe
##################################
mri_images_train['Image'] = mri_images_train['Path'].map(lambda x: np.asarray(Image.open(x).resize((200,200))))
In [32]:
##################################
# Listing the column names and data types
# for the training data
##################################
print('Column Names and Data Types:')
display(mri_images_train.dtypes)
Column Names and Data Types:
Image_ID object Path object Diagnosis object Target int64 Class object Image object dtype: object
In [33]:
##################################
# Taking a snapshot of the training data
##################################
mri_images_train.head()
Out[33]:
| Image_ID | Path | Diagnosis | Target | Class | Image | |
|---|---|---|---|---|---|---|
| 0 | Tr-glTr_0000 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
| 1 | Tr-glTr_0001 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
| 2 | Tr-glTr_0002 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
| 3 | Tr-glTr_0003 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
| 4 | Tr-glTr_0004 | ..\datasets\Brain_Tumor_MRI_Dataset\Training\g... | Tr-glTr | 1 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
In [34]:
##################################
# Taking a snapshot of the training data
##################################
n_samples = 5
fig, m_axs = plt.subplots(4, n_samples, figsize = (2*n_samples, 10))
for n_axs, (type_name, type_rows) in zip(m_axs, mri_images_train.sort_values(['Class']).groupby('Class')):
n_axs[2].set_title(type_name, fontsize = 14, weight = 'bold')
for c_ax, (_, c_row) in zip(n_axs, type_rows.sample(n_samples, random_state=123).iterrows()):
picture = c_row['Path']
image = cv2.imread(picture)
resized_image = cv2.resize(image, (500,500))
c_ax.imshow(resized_image)
c_ax.axis('off')
In [35]:
##################################
# Sampling a single image
# from the training data
##################################
samples, features = mri_images_train.shape
plt.figure()
pic_id = random.randrange(0, samples)
picture = mri_images_train['Path'][pic_id]
image = cv2.imread(picture)
<Figure size 640x480 with 0 Axes>
In [36]:
##################################
# Plotting using subplots
##################################
plt.figure(figsize=(15, 5))
##################################
# Formulating the original image
##################################
plt.subplot(1, 4, 1)
plt.imshow(image)
plt.title('Original Image', fontsize = 14, weight = 'bold')
plt.axis('off')
##################################
# Formulating the blue channel
##################################
plt.subplot(1, 4, 2)
plt.imshow(image[ : , : , 0])
plt.title('Blue Channel', fontsize = 14, weight = 'bold')
plt.axis('off')
##################################
# Formulating the green channel
##################################
plt.subplot(1, 4, 3)
plt.imshow(image[ : , : , 1])
plt.title('Green Channel', fontsize = 14, weight = 'bold')
plt.axis('off')
##################################
# Formulating the red channel
##################################
plt.subplot(1, 4, 4)
plt.imshow(image[ : , : , 2])
plt.title('Red Channel', fontsize = 14, weight = 'bold')
plt.axis('off')
##################################
# Consolidating all images
##################################
plt.show()
In [37]:
##################################
# Determining the image shape
##################################
print('Image Shape:')
display(image.shape)
Image Shape:
(512, 512, 3)
In [38]:
##################################
# Determining the image height
##################################
print('Image Height:')
display(image.shape[0])
Image Height:
512
In [39]:
##################################
# Determining the image width
##################################
print('Image Width:')
display(image.shape[1])
Image Width:
512
In [40]:
##################################
# Determining the image dimension
##################################
print('Image Dimension:')
display(image.ndim)
Image Dimension:
3
In [41]:
##################################
# Determining the image size
##################################
print('Image Size:')
display(image.size)
Image Size:
786432
In [42]:
##################################
# Determining the image data type
##################################
print('Image Data Type:')
display(image.dtype)
Image Data Type:
dtype('uint8')
In [43]:
##################################
# Determining the maximum RGB value
##################################
print('Image Maximum RGB:')
display(image.max())
Image Maximum RGB:
np.uint8(255)
In [44]:
##################################
# Determining the minimum RGB value
##################################
print('Image Minimum RGB:')
display(image.min())
Image Minimum RGB:
np.uint8(0)
In [45]:
##################################
# Identifying the path for the images
# and defining image categories
##################################
path_train = (os.path.join("..", DATASETS_FINAL_TRAIN_PATH))
classes=["notumor", "glioma", "meningioma", "pituitary"]
num_classes = len(classes)
batch_size = 32
In [46]:
##################################
# Creating subsets of images
# for model training and
# setting the parameters for
# real-time data augmentation
# at each epoch
##################################
set_seed()
train_datagen = ImageDataGenerator(rescale=1./255,
rotation_range=2,
width_shift_range=0.02,
height_shift_range=0.02,
horizontal_flip=False,
vertical_flip=False,
shear_range=0.02,
zoom_range=0.02,
validation_split=0.2)
##################################
# Loading the model training images
##################################
train_gen = train_datagen.flow_from_directory(directory=path_train,
target_size=(227, 227),
class_mode='categorical',
subset='training',
shuffle=True,
classes=classes,
batch_size=batch_size,
color_mode="grayscale")
Found 4571 images belonging to 4 classes.
In [47]:
##################################
# Loading samples of augmented images
# for the training set
##################################
fig, axes = plt.subplots(1, 5, figsize=(15, 3))
for i in range(5):
batch = next(train_gen)
images, labels = batch
axes[i].imshow(images[0])
axes[i].set_title(f"Label: {labels[0]}")
axes[i].axis('off')
plt.show()
In [48]:
##################################
# Creating subsets of images
# for model validation and
# setting the parameters for
# real-time data augmentation
# at each epoch
##################################
set_seed()
val_datagen = ImageDataGenerator(rescale=1./255,
validation_split=0.2)
##################################
# Loading the model evaluation images
##################################
val_gen = val_datagen.flow_from_directory(directory=path_train,
target_size=(227, 227),
class_mode='categorical',
subset='validation',
shuffle=False,
classes=classes,
batch_size=batch_size,
color_mode="grayscale")
Found 1141 images belonging to 4 classes.
In [49]:
##################################
# Loading samples of original images
# for the validation set
##################################
images, labels = next(val_gen)
fig, axes = plt.subplots(1, 5, figsize=(15, 3))
for i, idx in enumerate(range(0, 5)):
axes[i].imshow(images[idx])
axes[i].set_title(f"Label: {labels[0]}")
axes[i].axis('off')
plt.show()
In [50]:
##################################
# Including the pixel information
# of the actual images in array format
# for the testing data
# into a dataframe
##################################
mri_images_test['Image'] = mri_images_test['Path'].map(lambda x: np.asarray(Image.open(x).resize((200,200))))
In [51]:
##################################
# Listing the column names and data types
# for the testing data
##################################
print('Column Names and Data Types:')
display(mri_images_test.dtypes)
Column Names and Data Types:
Image_ID object Path object Diagnosis object Target float64 Class object Image object dtype: object
In [52]:
##################################
# Taking a snapshot of the testing data
##################################
mri_images_test.head()
Out[52]:
| Image_ID | Path | Diagnosis | Target | Class | Image | |
|---|---|---|---|---|---|---|
| 0 | Te-glTr_0000 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
| 1 | Te-glTr_0001 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
| 2 | Te-glTr_0002 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
| 3 | Te-glTr_0003 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
| 4 | Te-glTr_0004 | ..\datasets\Brain_Tumor_MRI_Dataset\Testing\gl... | Te-glTr | 1.0 | Glioma | [[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], ... |
In [53]:
##################################
# Taking a snapshot of the testing data
##################################
n_samples = 5
fig, m_axs = plt.subplots(4, n_samples, figsize = (2*n_samples, 10))
for n_axs, (type_name, type_rows) in zip(m_axs, mri_images_test.sort_values(['Class']).groupby('Class')):
n_axs[2].set_title(type_name, fontsize = 14, weight = 'bold')
for c_ax, (_, c_row) in zip(n_axs, type_rows.sample(n_samples, random_state=123).iterrows()):
picture = c_row['Path']
image = cv2.imread(picture)
resized_image = cv2.resize(image, (500,500))
c_ax.imshow(resized_image)
c_ax.axis('off')
In [54]:
##################################
# Sampling a single image
# from the testing data
##################################
samples, features = mri_images_test.shape
plt.figure()
pic_id = random.randrange(0, samples)
picture = mri_images_test['Path'][pic_id]
image = cv2.imread(picture)
<Figure size 640x480 with 0 Axes>
In [55]:
##################################
# Plotting using subplots
##################################
plt.figure(figsize=(15, 5))
##################################
# Formulating the original image
##################################
plt.subplot(1, 4, 1)
plt.imshow(image)
plt.title('Original Image', fontsize = 14, weight = 'bold')
plt.axis('off')
##################################
# Formulating the blue channel
##################################
plt.subplot(1, 4, 2)
plt.imshow(image[ : , : , 0])
plt.title('Blue Channel', fontsize = 14, weight = 'bold')
plt.axis('off')
##################################
# Formulating the green channel
##################################
plt.subplot(1, 4, 3)
plt.imshow(image[ : , : , 1])
plt.title('Green Channel', fontsize = 14, weight = 'bold')
plt.axis('off')
##################################
# Formulating the red channel
##################################
plt.subplot(1, 4, 4)
plt.imshow(image[ : , : , 2])
plt.title('Red Channel', fontsize = 14, weight = 'bold')
plt.axis('off')
##################################
# Consolidating all images
##################################
plt.show()
In [56]:
##################################
# Determining the image shape
##################################
print('Image Shape:')
display(image.shape)
Image Shape:
(512, 512, 3)
In [57]:
##################################
# Determining the image height
##################################
print('Image Height:')
display(image.shape[0])
Image Height:
512
In [58]:
##################################
# Determining the image width
##################################
print('Image Width:')
display(image.shape[1])
Image Width:
512
In [59]:
##################################
# Determining the image dimension
##################################
print('Image Dimension:')
display(image.ndim)
Image Dimension:
3
In [60]:
##################################
# Determining the image size
##################################
print('Image Size:')
display(image.size)
Image Size:
786432
In [61]:
##################################
# Determining the image data type
##################################
print('Image Data Type:')
display(image.dtype)
Image Data Type:
dtype('uint8')
In [62]:
##################################
# Determining the maximum RGB value
##################################
print('Image Maximum RGB:')
display(image.max())
Image Maximum RGB:
np.uint8(255)
In [63]:
##################################
# Determining the minimum RGB value
##################################
print('Image Minimum RGB:')
display(image.min())
Image Minimum RGB:
np.uint8(0)
In [64]:
##################################
# Identifying the path for the images
# and defining image categories
##################################
path_test = (os.path.join("..", DATASETS_FINAL_TEST_PATH))
classes=["notumor", "glioma", "meningioma", "pituitary"]
num_classes = len(classes)
batch_size = 32
In [65]:
##################################
# Creating subsets of images
# for model testing and
# setting the parameters for
# real-time data augmentation
# at each epoch
##################################
set_seed()
test_datagen = ImageDataGenerator(rescale=1./255)
##################################
# Loading the model testing images
##################################
test_gen = test_datagen.flow_from_directory(directory=path_test,
target_size=(227, 227),
class_mode='categorical',
shuffle=False,
classes=classes,
batch_size=batch_size,
color_mode="grayscale")
Found 1311 images belonging to 4 classes.
In [66]:
##################################
# Loading samples of augmented images
# for the testing set
##################################
fig, axes = plt.subplots(1, 5, figsize=(15, 3))
for i in range(5):
batch = next(test_gen)
images, labels = batch
axes[i].imshow(images[0])
axes[i].set_title(f"Label: {labels[0]}")
axes[i].axis('off')
plt.show()
1.5 Data Exploration ¶
1.5.1 Exploratory Data Analysis ¶
In [67]:
##################################
# Consolidating summary statistics
# for the image pixel values
##################################
samples, features = mri_images_train.shape
mean_val = []
std_dev_val = []
max_val = []
min_val = []
for i in range(0, samples):
mean_val.append(mri_images_train['Image'][i].mean())
std_dev_val.append(np.std(mri_images_train['Image'][i]))
max_val.append(mri_images_train['Image'][i].max())
min_val.append(mri_images_train['Image'][i].min())
imageEDA = mri_images_train.loc[:,['Image', 'Class','Path']]
imageEDA['Mean'] = mean_val
imageEDA['StDev'] = std_dev_val
imageEDA['Max'] = max_val
imageEDA['Min'] = min_val
In [68]:
##################################
# Consolidating the overall mean
# for the pixel intensity means
# grouped by categories
##################################
imageEDA.groupby(['Class'])['Mean'].mean()
Out[68]:
Class Glioma 32.716871 Meningioma 43.487954 No Tumor 60.815724 Pituitary 49.273456 Name: Mean, dtype: float64
In [69]:
##################################
# Consolidating the overall minimum
# for the pixel intensity means
# grouped by categories
##################################
imageEDA.groupby(['Class'])['Mean'].min()
Out[69]:
Class Glioma 13.701850 Meningioma 18.233400 No Tumor 9.770775 Pituitary 24.699575 Name: Mean, dtype: float64
In [70]:
##################################
# Consolidating the overall maximum
# for the pixel intensity means
# grouped by categories
##################################
imageEDA.groupby(['Class'])['Mean'].max()
Out[70]:
Class Glioma 68.372425 Meningioma 137.765375 No Tumor 125.066725 Pituitary 102.007950 Name: Mean, dtype: float64
In [71]:
##################################
# Consolidating the overall standard deviation
# for the pixel intensity means
# grouped by categories
##################################
imageEDA.groupby(['Class'])['Mean'].std()
Out[71]:
Class Glioma 8.565834 Meningioma 14.307165 No Tumor 21.338225 Pituitary 8.222902 Name: Mean, dtype: float64
In [72]:
##################################
# Formulating the mean distribution
# by category of the image pixel values
##################################
sns.displot(data = imageEDA, x = 'Mean', kind="kde", hue = 'Class', height=6, aspect=1.40)
plt.title('Image Pixel Intensity Mean Distribution by Category', fontsize=14, weight='bold');
In [73]:
##################################
# Formulating the maximum distribution
# by category of the image pixel values
##################################
sns.displot(data = imageEDA, x = 'Max', kind="kde", hue = 'Class', height=6, aspect=1.40)
plt.title('Image Pixel Intensity Maximum Distribution by Category', fontsize=14, weight='bold');
In [74]:
##################################
# Formulating the minimum distribution
# by category of the image pixel values
##################################
sns.displot(data = imageEDA, x = 'Min', kind="kde", hue = 'Class', height=6, aspect=1.40)
plt.title('Image Pixel Intensity Minimum Distribution by Category', fontsize=14, weight='bold');
In [75]:
##################################
# Formulating the standard deviation distribution
# by category of the image pixel values
##################################
sns.displot(data = imageEDA, x = 'StDev', kind="kde", hue = 'Class', height=6, aspect=1.40)
plt.title('Image Pixel Intensity Standard Deviation Distribution by Category', fontsize=14, weight='bold');
In [76]:
##################################
# Formulating the mean and standard deviation
# scatterplot distribution
# by category of the image pixel values
##################################
plt.figure(figsize=(10,6))
sns.set(style="ticks", font_scale = 1)
ax = sns.scatterplot(data=imageEDA, x="Mean", y=imageEDA['StDev'], hue='Class', alpha=0.5)
sns.despine(top=True, right=True, left=False, bottom=False)
plt.xticks(rotation=0, fontsize = 12)
ax.set_xlabel('Image Pixel Intensity Mean',fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
plt.title('Image Pixel Intensity Mean and Standard Deviation Distribution', fontsize = 14, weight='bold');
In [77]:
##################################
# Formulating the mean and standard deviation
# scatterplot distribution
# by category of the image pixel values
##################################
scatterplot = sns.FacetGrid(imageEDA, col="Class", height=6)
scatterplot.map_dataframe(sns.scatterplot, x='Mean', y='StDev', alpha=0.5)
scatterplot.set_titles(col_template="{col_name}", row_template="{row_name}", size=18)
scatterplot.fig.subplots_adjust(top=.8)
scatterplot.fig.suptitle('Image Pixel Intensity Mean and Standard Deviation Distribution', fontsize=14, weight='bold')
axes = scatterplot.axes.flatten()
axes[0].set_ylabel('Image Pixel Intensity Standard Deviation')
for ax in axes:
ax.set_xlabel('Image Pixel Intensity Mean')
scatterplot.fig.tight_layout()
In [78]:
##################################
# Formulating the mean and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
##################################
def getImage(path):
image = cv2.imread(path)
resized_image = cv2.resize(image, (300,300))
return OffsetImage(resized_image, zoom = 0.1)
DF_sample = imageEDA.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Mean", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Mean', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(-5,145)
ax.set_ylim(0,120)
plt.title('Overall: Image Pixel Intensity Mean and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path in zip(DF_sample['Mean'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path), (x0, y0), frameon=False)
ax.add_artist(ab)
In [79]:
##################################
# Formulating the mean and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the Glioma class
##################################
path_glioma = (os.path.join("..", DATASETS_FINAL_TRAIN_PATH,'glioma/'))
imageEDA_glioma = imageEDA.loc[imageEDA['Class'] == 'Glioma']
DF_sample = imageEDA_glioma.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Mean", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Mean', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(-5,145)
ax.set_ylim(10,110)
plt.title('Glioma: Image Pixel Intensity Mean and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_glioma in zip(DF_sample['Mean'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_glioma), (x0, y0), frameon=False)
ax.add_artist(ab)
In [80]:
##################################
# Formulating the mean and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the Viral Pneumonia class
##################################
path_meningioma = (os.path.join("..", DATASETS_FINAL_TRAIN_PATH,'meningioma/'))
imageEDA_meningioma = imageEDA.loc[imageEDA['Class'] == 'Meningioma']
DF_sample = imageEDA_meningioma.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Mean", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Mean', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(-5,145)
ax.set_ylim(10,110)
plt.title('Meningioma: Image Pixel Intensity Mean and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_meningioma in zip(DF_sample['Mean'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_meningioma), (x0, y0), frameon=False)
ax.add_artist(ab)
In [81]:
##################################
# Formulating the mean and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the Pituitary class
##################################
path_pituitary = (os.path.join("..", DATASETS_FINAL_TRAIN_PATH,'pituitary/'))
imageEDA_pituitary = imageEDA.loc[imageEDA['Class'] == 'Pituitary']
DF_sample = imageEDA_pituitary.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Mean", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Mean', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(0, 140)
ax.set_ylim(10,110)
plt.title('Pituitary: Image Pixel Intensity Mean and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_pituitary in zip(DF_sample['Mean'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_pituitary), (x0, y0), frameon=False)
ax.add_artist(ab)
In [82]:
##################################
# Formulating the mean and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the No Tumor class
##################################
path_no_tumor = (os.path.join("..", DATASETS_FINAL_TRAIN_PATH,'notumor/'))
imageEDA_no_tumor = imageEDA.loc[imageEDA['Class'] == 'No Tumor']
DF_sample = imageEDA_no_tumor.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Mean", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Mean', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(-5,145)
ax.set_ylim(10,110)
plt.title('No Tumor: Image Pixel Intensity Mean and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_no_tumor in zip(DF_sample['Mean'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_no_tumor), (x0, y0), frameon=False)
ax.add_artist(ab)
In [83]:
#################################
# Formulating the minimum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
##################################
DF_sample = imageEDA.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Min", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Minimum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(-5,145)
ax.set_ylim(0,120)
plt.title('Overall: Image Pixel Intensity Minimum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path in zip(DF_sample['Min'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path), (x0, y0), frameon=False)
ax.add_artist(ab)
In [84]:
##################################
# Formulating the minimum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the Glioma class
##################################
DF_sample = imageEDA_glioma.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Min", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Minimum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(-5,145)
ax.set_ylim(10,110)
plt.title('Glioma: Image Pixel Intensity Minimum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_glioma in zip(DF_sample['Min'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_glioma), (x0, y0), frameon=False)
ax.add_artist(ab)
In [85]:
##################################
# Formulating the minimum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the Meningioma class
##################################
DF_sample = imageEDA_meningioma.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Min", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Minimum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(-5,145)
ax.set_ylim(10,110)
plt.title('Meningioma: Image Pixel Intensity Minimum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_meningioma in zip(DF_sample['Min'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_meningioma), (x0, y0), frameon=False)
ax.add_artist(ab)
In [86]:
##################################
# Formulating the minimum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the Pituitary class
##################################
DF_sample = imageEDA_pituitary.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Min", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Minimum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(-5,145)
ax.set_ylim(10,110)
plt.title('Pituitary: Image Pixel Intensity Minimum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_pituitary in zip(DF_sample['Min'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_pituitary), (x0, y0), frameon=False)
ax.add_artist(ab)
In [87]:
##################################
# Formulating the minimum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the No Tumor class
##################################
DF_sample = imageEDA_no_tumor.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Min", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Minimum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(-5,145)
ax.set_ylim(10,110)
plt.title('No Tumor: Image Pixel Intensity Minimum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_no_tumor in zip(DF_sample['Min'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_no_tumor), (x0, y0), frameon=False)
ax.add_artist(ab)
In [88]:
#################################
# Formulating the maximum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
##################################
DF_sample = imageEDA.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Max", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Maximum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(115,265)
ax.set_ylim(0,120)
plt.title('Overall: Image Pixel Intensity Maximum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path in zip(DF_sample['Max'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path), (x0, y0), frameon=False)
ax.add_artist(ab)
In [89]:
##################################
# Formulating the maximum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the Glioma class
##################################
DF_sample = imageEDA_glioma.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Max", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Maximum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(115,265)
ax.set_ylim(10,110)
plt.title('Glioma: Image Pixel Intensity Maximum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_glioma in zip(DF_sample['Max'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_glioma), (x0, y0), frameon=False)
ax.add_artist(ab)
In [90]:
##################################
# Formulating the maximum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the Meningioma class
##################################
DF_sample = imageEDA_meningioma.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Max", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Maximum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(115,265)
ax.set_ylim(10,110)
plt.title('Meningioma: Image Pixel Intensity Maximum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_meningioma in zip(DF_sample['Max'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_meningioma), (x0, y0), frameon=False)
ax.add_artist(ab)
In [91]:
##################################
# Formulating the maximum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the Pituitary class
##################################
DF_sample = imageEDA_pituitary.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Max", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Maximum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(115,265)
ax.set_ylim(10,110)
plt.title('Pituitary: Image Pixel Intensity Maximum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_pituitary in zip(DF_sample['Max'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_pituitary), (x0, y0), frameon=False)
ax.add_artist(ab)
In [92]:
##################################
# Formulating the maximum and standard deviation
# scatterplot distribution
# of the image pixel values
# represented as actual images
# for the No Tumor class
##################################
DF_sample = imageEDA_no_tumor.sample(frac=1.0, replace=False, random_state=123)
paths = DF_sample['Path']
fig, ax = plt.subplots(figsize=(15,9))
ab = sns.scatterplot(data=DF_sample, x="Max", y='StDev')
sns.despine(top=True, right=True, left=False, bottom=False)
ax.set_xlabel('Image Pixel Intensity Maximum', fontsize=14, weight='bold')
ax.set_ylabel('Image Pixel Intensity Standard Deviation', fontsize=14, weight='bold')
ax.set_xlim(115,265)
ax.set_ylim(10,110)
plt.title('No Tumor: Image Pixel Intensity Maximum and Standard Deviation Distribution', fontsize=14, weight='bold');
for x0, y0, path_no_tumor in zip(DF_sample['Max'], DF_sample['StDev'], paths):
ab = AnnotationBbox(getImage(path_no_tumor), (x0, y0), frameon=False)
ax.add_artist(ab)
1.5.2 Hypothesis Testing ¶
1.6 Predictive Model Development ¶
1.6.1 Pre-Modelling Data Preparation ¶
1.6.2 Convolutional Neural Network Sequential Layer Development ¶
- The convolutional neural network model from the keras.models Python library API was implemented using a sequential model structure.
- Specialized functions were applied to fine-tune the training process dynamically and automate certain actions including:
- EarlyStopping to stop training when a monitored metric stops improving which helps to prevent overfitting and saves time, with fixed hyperparameters as follows:
- monitor = validation loss (metric to track)
- patience = 10 (number of epochs with no improvement after which training stops)
- restore_best_weights = true (model's weights will be restored to those from the epoch with the best value of the monitored metric)
- min_delta = 0.0001 (minimum change in the monitored metric to qualify as an improvement)
- ReduceLROnPlateau to reduce the learning rate when a monitored metric has stopped improving which helps the model converge better by fine-tuning updates, with fixed hyperparameters as follows:
- monitor = validation loss (metric to track)
- factor = 0.10 (percentage by which the learning rate is reduced)
- patience = 3 (number of epochs with no improvement after which training stops)
- min_lr = 0.000001 (lower bound for the learning rate to prevent it from becoming too small)
- ModelCheckpoint to save the model at specified intervals during training and enabling the sunsequent restoration of the best-performing model, with fixed hyperparameters as follows:
- monitor = validation loss (metric to track)
- save_best_only = true (only saves the model when the monitored metric improves)
- save_weights_only = false (saving the entire model and not just the weights)
- EarlyStopping to stop training when a monitored metric stops improving which helps to prevent overfitting and saves time, with fixed hyperparameters as follows:
In [93]:
##################################
# Defining a function for
# plotting the loss profile
# of the training and validation sets
#################################
def plot_training_history(history, model_name):
plt.figure(figsize=(12, 8))
# Plotting training and validation loss
plt.subplot(2, 1, 1) # First subplot for loss
plt.plot(history.history['loss'], label='Train Loss', color='blue')
plt.plot(history.history['val_loss'], label='Validation Loss', color='orange')
plt.title(f'{model_name} Training and Validation Loss', fontsize=16, weight='bold', pad=20)
plt.ylim(-0.2, 2.2)
plt.yticks([x * 0.50 for x in range(0, 5)])
plt.xlim(-1, 21)
plt.xticks([x for x in range(0, 21)])
plt.xlabel('Epoch', fontsize=14, weight='bold')
plt.ylabel('Loss', fontsize=14, weight='bold')
plt.legend(loc='upper right')
plt.grid(True)
# Plotting training and validation recall
plt.subplot(2, 1, 2) # Second subplot for recall
plt.plot(history.history['recall'], label='Train Recall', color='green')
plt.plot(history.history['val_recall'], label='Validation Recall', color='red')
plt.title(f'{model_name} Training and Validation Recall', fontsize=16, weight='bold', pad=20)
plt.ylim(-0.1, 1.1)
plt.yticks([x * 0.25 for x in range(0, 5)])
plt.xlim(-1, 21)
plt.xticks([x for x in range(0, 21)])
plt.xlabel('Epoch', fontsize=14, weight='bold')
plt.ylabel('Recall', fontsize=14, weight='bold')
plt.legend(loc='lower right')
plt.grid(True)
# Adjusting layout and show the plots
plt.tight_layout(pad=2.0)
plt.show()
In [94]:
##################################
# Defining the model file paths
#################################
NR_SIMPLE_BEST_MODEL_PATH = os.path.join("..", MODELS_PATH, "nr_simple_best_model.keras")
DR_SIMPLE_BEST_MODEL_PATH = os.path.join("..", MODELS_PATH, "dr_simple_best_model.keras")
BNR_SIMPLE_BEST_MODEL_PATH = os.path.join("..", MODELS_PATH, "bnr_simple_best_model.keras")
CDRBNR_SIMPLE_BEST_MODEL_PATH = os.path.join("..", MODELS_PATH, "cdrbnr_simple_best_model.keras")
NR_COMPLEX_BEST_MODEL_PATH = os.path.join("..", MODELS_PATH, "nr_complex_best_model.keras")
DR_COMPLEX_BEST_MODEL_PATH = os.path.join("..", MODELS_PATH, "dr_complex_best_model.keras")
BNR_COMPLEX_BEST_MODEL_PATH = os.path.join("..", MODELS_PATH, "bnr_complex_best_model.keras")
CDRBNR_COMPLEX_BEST_MODEL_PATH = os.path.join("..", MODELS_PATH, "cdrbnr_complex_best_model.keras")
In [95]:
##################################
# Defining the model callback configuration
# for model training
#################################
early_stopping = EarlyStopping(
monitor='val_loss', # Defining the metric to monitor
patience=10, # Defining the number of epochs to wait before stopping if no improvement
min_delta=1e-4, # Defining the minimum change in the monitored quantity to qualify as an improvement
restore_best_weights=True # Restoring the weights from the best epoch
)
reduce_lr = ReduceLROnPlateau(
monitor='val_loss', # Defining the metric to monitor
factor=0.1, # Reducing the learning rate by a factor of 10%
patience=3, # Defining the number of epochs to wait before reducing learning rate
min_lr=1e-6 # Defining the lower bound on the learning rate
)
nr_simple_model_checkpoint = ModelCheckpoint(
filepath=NR_SIMPLE_BEST_MODEL_PATH, # Defining the file path for saving
monitor='val_loss', # Defining the metric to monitor
save_best_only=True, # Saving only the best model
save_weights_only=False, # Saving the entire model, not just weights
)
dr_simple_model_checkpoint = ModelCheckpoint(
filepath=DR_SIMPLE_BEST_MODEL_PATH, # Defining the file path for saving
monitor='val_loss', # Defining the metric to monitor
save_best_only=True, # Saving only the best model
save_weights_only=False, # Saving the entire model, not just weights
)
bnr_simple_model_checkpoint = ModelCheckpoint(
filepath=BNR_SIMPLE_BEST_MODEL_PATH, # Defining the file path for saving
monitor='val_loss', # Defining the metric to monitor
save_best_only=True, # Saving only the best model
save_weights_only=False, # Saving the entire model, not just weights
)
cdrbnr_simple_model_checkpoint = ModelCheckpoint(
filepath=CDRBNR_SIMPLE_BEST_MODEL_PATH, # Defining the file path for saving
monitor='val_loss', # Defining the metric to monitor
save_best_only=True, # Saving only the best model
save_weights_only=False, # Saving the entire model, not just weights
)
nr_complex_model_checkpoint = ModelCheckpoint(
filepath=NR_COMPLEX_BEST_MODEL_PATH, # Defining the file path for saving
monitor='val_loss', # Defining the metric to monitor
save_best_only=True, # Saving only the best model
save_weights_only=False, # Saving the entire model, not just weights
)
dr_complex_model_checkpoint = ModelCheckpoint(
filepath=DR_COMPLEX_BEST_MODEL_PATH, # Defining the file path for saving
monitor='val_loss', # Defining the metric to monitor
save_best_only=True, # Saving only the best model
save_weights_only=False, # Saving the entire model, not just weights
)
bnr_complex_model_checkpoint = ModelCheckpoint(
filepath=BNR_COMPLEX_BEST_MODEL_PATH, # Defining the file path for saving
monitor='val_loss', # Defining the metric to monitor
save_best_only=True, # Saving only the best model
save_weights_only=False, # Saving the entire model, not just weights
)
cdrbnr_complex_model_checkpoint = ModelCheckpoint(
filepath=CDRBNR_COMPLEX_BEST_MODEL_PATH, # Defining the file path for saving
monitor='val_loss', # Defining the metric to monitor
save_best_only=True, # Saving only the best model
save_weights_only=False, # Saving the entire model, not just weights
)
1.6.2.1 CNN With No Regularization ¶
- The simple model contains 7 layers with fixed hyperparameters as follows:
- Conv2D: nr_simple_conv2d_0
- filters = 8
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- input_shape = 227x227x1
- MaxPooling2D: nr_simple_max_pooling2d_0
- pool_size = 2x2
- Conv2D: nr_simple_conv2d_1
- filters = 16
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- MaxPooling2D: nr_simple_max_pooling2d_1
- pool_size = 2x2
- Flatten: nr_simple_flatten
- Dense: nr_simple_dense_0
- units = 32
- activation = relu (rectified linear unit)
- Dense: nr_simple_dense_1
- units = 4
- activation = softmax
- Conv2D: nr_simple_conv2d_0
- The complex model contains 9 layers with fixed hyperparameters as follows:
- Conv2D: nr_complex_conv2d_0
- filters = 16
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- input_shape = 227x227x1
- MaxPooling2D: nr_complex_max_pooling2d_0
- pool_size = 2x2
- Conv2D: nr_complex_conv2d_1
- filters = 32
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- MaxPooling2D: nr_complex_max_pooling2d_1
- pool_size = 2x2
- Conv2D: nr_complex_conv2d_2
- filters = 64
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- MaxPooling2D: nr_complex_max_pooling2d_2
- pool_size = 2x2
- Flatten: nr_complex_flatten
- Dense: nr_complex_dense_0
- units = 128
- activation = relu (rectified linear unit)
- Dense: nr_complex_dense_1
- units = 4
- activation = softmax
- Conv2D: nr_complex_conv2d_0
- Additional fixed hyperparameters used during model compilation are as follows:
- loss = categorical_crossentropy
- optimizer = adam (adaptive moment estimation)
- metrics = recall
In [96]:
##################################
# Formulating the network architecture
# for a simple CNN with no regularization
##################################
set_seed()
batch_size = 32
model_nr_simple = Sequential(name="model_nr_simple")
model_nr_simple.add(Conv2D(filters=8, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="nr_simple_conv2d_0"))
model_nr_simple.add(MaxPooling2D(pool_size=(2, 2), name="nr_simple_max_pooling2d_0"))
model_nr_simple.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', name="nr_simple_conv2d_1"))
model_nr_simple.add(MaxPooling2D(pool_size=(2, 2), name="nr_simple_max_pooling2d_1"))
model_nr_simple.add(Flatten(name="nr_simple_flatten"))
model_nr_simple.add(Dense(units=32, activation='relu', name="nr_simple_dense_0"))
model_nr_simple.add(Dense(units=num_classes, activation='softmax', name="nr_simple_dense_1"))
In [97]:
##################################
# Displaying the model summary
# for a simple CNN with no regularization
##################################
print(model_nr_simple.summary())
Model: "model_nr_simple"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ nr_simple_conv2d_0 (Conv2D) │ (None, 227, 227, 8) │ 80 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_simple_max_pooling2d_0 │ (None, 113, 113, 8) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_simple_conv2d_1 (Conv2D) │ (None, 113, 113, 16) │ 1,168 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_simple_max_pooling2d_1 │ (None, 56, 56, 16) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_simple_flatten (Flatten) │ (None, 50176) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_simple_dense_0 (Dense) │ (None, 32) │ 1,605,664 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_simple_dense_1 (Dense) │ (None, 4) │ 132 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 1,607,044 (6.13 MB)
Trainable params: 1,607,044 (6.13 MB)
Non-trainable params: 0 (0.00 B)
None
In [98]:
##################################
# Displaying the model layers
# for a simple CNN with no regularization
##################################
model_nr_simple_layer_names = [layer.name for layer in model_nr_simple.layers]
print("Layer Names:", model_nr_simple_layer_names)
Layer Names: ['nr_simple_conv2d_0', 'nr_simple_max_pooling2d_0', 'nr_simple_conv2d_1', 'nr_simple_max_pooling2d_1', 'nr_simple_flatten', 'nr_simple_dense_0', 'nr_simple_dense_1']
In [99]:
##################################
# Displaying the number of weights
# for each model layer
# for a simple CNN with no regularization
##################################
for layer in model_nr_simple.layers:
if hasattr(layer, 'weights'):
print(f"Layer: {layer.name}, Number of Weights: {len(layer.get_weights())}")
Layer: nr_simple_conv2d_0, Number of Weights: 2 Layer: nr_simple_max_pooling2d_0, Number of Weights: 0 Layer: nr_simple_conv2d_1, Number of Weights: 2 Layer: nr_simple_max_pooling2d_1, Number of Weights: 0 Layer: nr_simple_flatten, Number of Weights: 0 Layer: nr_simple_dense_0, Number of Weights: 2 Layer: nr_simple_dense_1, Number of Weights: 2
In [100]:
##################################
# Displaying the number of parameters
# for each model layer
# for a simple CNN with no regularization
##################################
total_parameters = 0
for layer in model_nr_simple.layers:
layer_parameters = layer.count_params()
total_parameters += layer_parameters
print(f"Layer: {layer.name}, Parameters: {layer_parameters}")
print("\nTotal Parameters in the Model:", total_parameters)
Layer: nr_simple_conv2d_0, Parameters: 80 Layer: nr_simple_max_pooling2d_0, Parameters: 0 Layer: nr_simple_conv2d_1, Parameters: 1168 Layer: nr_simple_max_pooling2d_1, Parameters: 0 Layer: nr_simple_flatten, Parameters: 0 Layer: nr_simple_dense_0, Parameters: 1605664 Layer: nr_simple_dense_1, Parameters: 132 Total Parameters in the Model: 1607044
In [101]:
##################################
# Formulating the network architecture
# for a complex CNN with no regularization
##################################
set_seed()
batch_size = 32
model_nr_complex = Sequential(name="model_nr_complex")
model_nr_complex.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="nr_complex_conv2d_0"))
model_nr_complex.add(MaxPooling2D(pool_size=(2, 2), name="nr_complex_max_pooling2d_0"))
model_nr_complex.add(Conv2D(filters=32, kernel_size=(3, 3), padding = 'Same', activation='relu', name="nr_complex_conv2d_1"))
model_nr_complex.add(MaxPooling2D(pool_size=(2, 2), name="nr_complex_max_pooling2d_1"))
model_nr_complex.add(Conv2D(filters=64, kernel_size=(3, 3), padding = 'Same', activation='relu', name="nr_complex_conv2d_2"))
model_nr_complex.add(MaxPooling2D(pool_size=(2, 2), name="nr_complex_max_pooling2d_2"))
model_nr_complex.add(Flatten(name="nr_complex_flatten"))
model_nr_complex.add(Dense(units=128, activation='relu', name="nr_complex_dense_0"))
model_nr_complex.add(Dense(units=num_classes, activation='softmax', name="nr_complex_dense_1"))
In [102]:
##################################
# Displaying the model summary
# for a complex CNN with no regularization
##################################
print(model_nr_complex.summary())
Model: "model_nr_complex"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ nr_complex_conv2d_0 (Conv2D) │ (None, 227, 227, 16) │ 160 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_complex_max_pooling2d_0 │ (None, 113, 113, 16) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_complex_conv2d_1 (Conv2D) │ (None, 113, 113, 32) │ 4,640 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_complex_max_pooling2d_1 │ (None, 56, 56, 32) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_complex_conv2d_2 (Conv2D) │ (None, 56, 56, 64) │ 18,496 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_complex_max_pooling2d_2 │ (None, 28, 28, 64) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_complex_flatten (Flatten) │ (None, 50176) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_complex_dense_0 (Dense) │ (None, 128) │ 6,422,656 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ nr_complex_dense_1 (Dense) │ (None, 4) │ 516 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 6,446,468 (24.59 MB)
Trainable params: 6,446,468 (24.59 MB)
Non-trainable params: 0 (0.00 B)
None
In [103]:
##################################
# Displaying the model layers
# for a complex CNN with no regularization
##################################
model_nr_complex_layer_names = [layer.name for layer in model_nr_complex.layers]
print("Layer Names:", model_nr_complex_layer_names)
Layer Names: ['nr_complex_conv2d_0', 'nr_complex_max_pooling2d_0', 'nr_complex_conv2d_1', 'nr_complex_max_pooling2d_1', 'nr_complex_conv2d_2', 'nr_complex_max_pooling2d_2', 'nr_complex_flatten', 'nr_complex_dense_0', 'nr_complex_dense_1']
In [104]:
##################################
# Displaying the number of weights
# for each model layer
# for a complex CNN with no regularization
##################################
for layer in model_nr_complex.layers:
if hasattr(layer, 'weights'):
print(f"Layer: {layer.name}, Number of Weights: {len(layer.get_weights())}")
Layer: nr_complex_conv2d_0, Number of Weights: 2 Layer: nr_complex_max_pooling2d_0, Number of Weights: 0 Layer: nr_complex_conv2d_1, Number of Weights: 2 Layer: nr_complex_max_pooling2d_1, Number of Weights: 0 Layer: nr_complex_conv2d_2, Number of Weights: 2 Layer: nr_complex_max_pooling2d_2, Number of Weights: 0 Layer: nr_complex_flatten, Number of Weights: 0 Layer: nr_complex_dense_0, Number of Weights: 2 Layer: nr_complex_dense_1, Number of Weights: 2
In [105]:
##################################
# Displaying the number of parameters
# for each model layer
# for a complex CNN with no regularization
##################################
total_parameters = 0
for layer in model_nr_complex.layers:
layer_parameters = layer.count_params()
total_parameters += layer_parameters
print(f"Layer: {layer.name}, Parameters: {layer_parameters}")
print("\nTotal Parameters in the Model:", total_parameters)
Layer: nr_complex_conv2d_0, Parameters: 160 Layer: nr_complex_max_pooling2d_0, Parameters: 0 Layer: nr_complex_conv2d_1, Parameters: 4640 Layer: nr_complex_max_pooling2d_1, Parameters: 0 Layer: nr_complex_conv2d_2, Parameters: 18496 Layer: nr_complex_max_pooling2d_2, Parameters: 0 Layer: nr_complex_flatten, Parameters: 0 Layer: nr_complex_dense_0, Parameters: 6422656 Layer: nr_complex_dense_1, Parameters: 516 Total Parameters in the Model: 6446468
1.6.2.2 CNN With Dropout Regularization ¶
- The simple model contains 8 layers with fixed hyperparameters as follows:
- Conv2D: dr_simple_conv2d_0
- filters = 8
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- input_shape = 227x227x1
- MaxPooling2D: dr_simple_max_pooling2d_0
- pool_size = 2x2
- Conv2D: dr_simple_conv2d_1
- filters = 16
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- MaxPooling2D: dr_simple_max_pooling2d_1
- pool_size = 2x2
- Flatten: dr_simple_flatten
- Dense: dr_simple_dense_0
- units = 32
- activation = relu (rectified linear unit)
- Dropout: dr_simple_dropout
- rate = 0.25
- Dense: dr_simple_dense_1
- units = 4
- activation = softmax
- Conv2D: dr_simple_conv2d_0
- The complex model contains 10 layers with fixed hyperparameters as follows:
- Conv2D: dr_complex_conv2d_0
- filters = 16
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- input_shape = 227x227x1
- MaxPooling2D: dr_complex_max_pooling2d_0
- pool_size = 2x2
- Conv2D: dr_complex_conv2d_1
- filters = 32
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- MaxPooling2D: dr_complex_max_pooling2d_1
- pool_size = 2x2
- Conv2D: dr_complex_conv2d_2
- filters = 64
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- MaxPooling2D: dr_complex_max_pooling2d_2
- pool_size = 2x2
- Flatten: dr_complex_flatten
- Dense: dr_complex_dense_0
- units = 128
- activation = relu (rectified linear unit)
- Dropout: dr_complex_dropout
- rate = 0.25
- Dense: dr_complex_dense_1
- units = 4
- activation = softmax
- Conv2D: dr_complex_conv2d_0
- Additional fixed hyperparameters used during model compilation are as follows:
- loss = categorical_crossentropy
- optimizer = adam (adaptive moment estimation)
- metrics = recall
In [106]:
##################################
# Formulating the network architecture
# for a simple CNN with dropout regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_dr_simple = Sequential(name="model_dr_simple")
model_dr_simple.add(Conv2D(filters=8, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="dr_simple_conv2d_0"))
model_dr_simple.add(MaxPooling2D(pool_size=(2, 2), name="dr_simple_max_pooling2d_0"))
model_dr_simple.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', name="dr_simple_conv2d_1"))
model_dr_simple.add(MaxPooling2D(pool_size=(2, 2), name="dr_simple_max_pooling2d_1"))
model_dr_simple.add(Flatten(name="dr_simple_flatten"))
model_dr_simple.add(Dense(units=32, activation='relu', name="dr_simple_dense_0"))
model_dr_simple.add(Dropout(rate=0.30, name="dr_simple_dropout"))
model_dr_simple.add(Dense(units=num_classes, activation='softmax', name="dr_simple_dense_1"))
In [107]:
##################################
# Displaying the model summary
# for a simple CNN with dropout regularization
##################################
print(model_dr_simple.summary())
Model: "model_dr_simple"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ dr_simple_conv2d_0 (Conv2D) │ (None, 227, 227, 8) │ 80 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_simple_max_pooling2d_0 │ (None, 113, 113, 8) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_simple_conv2d_1 (Conv2D) │ (None, 113, 113, 16) │ 1,168 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_simple_max_pooling2d_1 │ (None, 56, 56, 16) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_simple_flatten (Flatten) │ (None, 50176) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_simple_dense_0 (Dense) │ (None, 32) │ 1,605,664 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_simple_dropout (Dropout) │ (None, 32) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_simple_dense_1 (Dense) │ (None, 4) │ 132 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 1,607,044 (6.13 MB)
Trainable params: 1,607,044 (6.13 MB)
Non-trainable params: 0 (0.00 B)
None
In [108]:
##################################
# Displaying the model layers
# for a simple CNN with dropout regularization
##################################
model_dr_simple_layer_names = [layer.name for layer in model_dr_simple.layers]
print("Layer Names:", model_dr_simple_layer_names)
Layer Names: ['dr_simple_conv2d_0', 'dr_simple_max_pooling2d_0', 'dr_simple_conv2d_1', 'dr_simple_max_pooling2d_1', 'dr_simple_flatten', 'dr_simple_dense_0', 'dr_simple_dropout', 'dr_simple_dense_1']
In [109]:
##################################
# Displaying the number of weights
# for each model layer
# for a simple CNN with dropout regularization
##################################
for layer in model_dr_simple.layers:
if hasattr(layer, 'weights'):
print(f"Layer: {layer.name}, Number of Weights: {len(layer.get_weights())}")
Layer: dr_simple_conv2d_0, Number of Weights: 2 Layer: dr_simple_max_pooling2d_0, Number of Weights: 0 Layer: dr_simple_conv2d_1, Number of Weights: 2 Layer: dr_simple_max_pooling2d_1, Number of Weights: 0 Layer: dr_simple_flatten, Number of Weights: 0 Layer: dr_simple_dense_0, Number of Weights: 2 Layer: dr_simple_dropout, Number of Weights: 0 Layer: dr_simple_dense_1, Number of Weights: 2
In [110]:
##################################
# Displaying the number of parameters
# for each model layer
# for a simple CNN with dropout regularization
##################################
total_parameters = 0
for layer in model_dr_simple.layers:
layer_parameters = layer.count_params()
total_parameters += layer_parameters
print(f"Layer: {layer.name}, Parameters: {layer_parameters}")
print("\nTotal Parameters in the Model:", total_parameters)
Layer: dr_simple_conv2d_0, Parameters: 80 Layer: dr_simple_max_pooling2d_0, Parameters: 0 Layer: dr_simple_conv2d_1, Parameters: 1168 Layer: dr_simple_max_pooling2d_1, Parameters: 0 Layer: dr_simple_flatten, Parameters: 0 Layer: dr_simple_dense_0, Parameters: 1605664 Layer: dr_simple_dropout, Parameters: 0 Layer: dr_simple_dense_1, Parameters: 132 Total Parameters in the Model: 1607044
In [111]:
##################################
# Formulating the network architecture
# for a complex CNN with dropout regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_dr_complex = Sequential(name="model_dr_complex")
model_dr_complex.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="dr_complex_conv2d_0"))
model_dr_complex.add(MaxPooling2D(pool_size=(2, 2), name="dr_complex_max_pooling2d_0"))
model_dr_complex.add(Conv2D(filters=32, kernel_size=(3, 3), padding = 'Same', activation='relu', name="dr_complex_conv2d_1"))
model_dr_complex.add(MaxPooling2D(pool_size=(2, 2), name="dr_complex_max_pooling2d_1"))
model_dr_complex.add(Conv2D(filters=64, kernel_size=(3, 3), padding = 'Same', activation='relu', name="dr_complex_conv2d_2"))
model_dr_complex.add(MaxPooling2D(pool_size=(2, 2), name="dr_complex_max_pooling2d_2"))
model_dr_complex.add(Flatten(name="dr_complex_flatten"))
model_dr_complex.add(Dense(units=128, activation='relu', name="dr_complex_dense_0"))
model_dr_complex.add(Dropout(rate=0.30, name="dr_complex_dropout"))
model_dr_complex.add(Dense(units=num_classes, activation='softmax', name="dr_complex_dense_1"))
In [112]:
##################################
# Displaying the model summary
# for a complex CNN with dropout regularization
##################################
print(model_dr_complex.summary())
Model: "model_dr_complex"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ dr_complex_conv2d_0 (Conv2D) │ (None, 227, 227, 16) │ 160 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_complex_max_pooling2d_0 │ (None, 113, 113, 16) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_complex_conv2d_1 (Conv2D) │ (None, 113, 113, 32) │ 4,640 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_complex_max_pooling2d_1 │ (None, 56, 56, 32) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_complex_conv2d_2 (Conv2D) │ (None, 56, 56, 64) │ 18,496 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_complex_max_pooling2d_2 │ (None, 28, 28, 64) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_complex_flatten (Flatten) │ (None, 50176) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_complex_dense_0 (Dense) │ (None, 128) │ 6,422,656 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_complex_dropout (Dropout) │ (None, 128) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dr_complex_dense_1 (Dense) │ (None, 4) │ 516 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 6,446,468 (24.59 MB)
Trainable params: 6,446,468 (24.59 MB)
Non-trainable params: 0 (0.00 B)
None
In [113]:
##################################
# Displaying the model layers
# for a complex CNN with dropout regularization
##################################
model_dr_complex_layer_names = [layer.name for layer in model_dr_complex.layers]
print("Layer Names:", model_dr_complex_layer_names)
Layer Names: ['dr_complex_conv2d_0', 'dr_complex_max_pooling2d_0', 'dr_complex_conv2d_1', 'dr_complex_max_pooling2d_1', 'dr_complex_conv2d_2', 'dr_complex_max_pooling2d_2', 'dr_complex_flatten', 'dr_complex_dense_0', 'dr_complex_dropout', 'dr_complex_dense_1']
In [114]:
##################################
# Displaying the number of weights
# for each model layer
# for a complex CNN with dropout regularization
##################################
for layer in model_dr_complex.layers:
if hasattr(layer, 'weights'):
print(f"Layer: {layer.name}, Number of Weights: {len(layer.get_weights())}")
Layer: dr_complex_conv2d_0, Number of Weights: 2 Layer: dr_complex_max_pooling2d_0, Number of Weights: 0 Layer: dr_complex_conv2d_1, Number of Weights: 2 Layer: dr_complex_max_pooling2d_1, Number of Weights: 0 Layer: dr_complex_conv2d_2, Number of Weights: 2 Layer: dr_complex_max_pooling2d_2, Number of Weights: 0 Layer: dr_complex_flatten, Number of Weights: 0 Layer: dr_complex_dense_0, Number of Weights: 2 Layer: dr_complex_dropout, Number of Weights: 0 Layer: dr_complex_dense_1, Number of Weights: 2
In [115]:
##################################
# Displaying the number of parameters
# for each model layer
# for a complex CNN with dropout regularization
##################################
total_parameters = 0
for layer in model_dr_complex.layers:
layer_parameters = layer.count_params()
total_parameters += layer_parameters
print(f"Layer: {layer.name}, Parameters: {layer_parameters}")
print("\nTotal Parameters in the Model:", total_parameters)
Layer: dr_complex_conv2d_0, Parameters: 160 Layer: dr_complex_max_pooling2d_0, Parameters: 0 Layer: dr_complex_conv2d_1, Parameters: 4640 Layer: dr_complex_max_pooling2d_1, Parameters: 0 Layer: dr_complex_conv2d_2, Parameters: 18496 Layer: dr_complex_max_pooling2d_2, Parameters: 0 Layer: dr_complex_flatten, Parameters: 0 Layer: dr_complex_dense_0, Parameters: 6422656 Layer: dr_complex_dropout, Parameters: 0 Layer: dr_complex_dense_1, Parameters: 516 Total Parameters in the Model: 6446468
1.6.2.3 CNN With Batch Normalization Regularization ¶
- The simple model contains 9 layers with fixed hyperparameters as follows:
- Conv2D: bnr_simple_conv2d_0
- filters = 8
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- input_shape = 227x227x1
- MaxPooling2D: bnr_simple_max_pooling2d_0
- pool_size = 2x2
- Conv2D: bnr_simple_conv2d_1
- filters = 16
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- BatchNormalization: bnr_simple_batch_normalization
- Activation: bnr_simple_activation
- activation = relu (rectified linear unit)
- MaxPooling2D: bnr_simple_max_pooling2d_1
- pool_size = 2x2
- Flatten: bnr_simple_flatten
- Dense: bnr_simple_dense_0
- units = 32
- activation = relu (rectified linear unit)
- Dense: bnr_simple_dense_1
- units = 4
- activation = softmax
- Conv2D: bnr_simple_conv2d_0
- The complex model contains 10 layers with fixed hyperparameters as follows:
- Conv2D: bnr_complex_conv2d_0
- filters = 16
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- input_shape = 227x227x1
- MaxPooling2D: bnr_complex_max_pooling2d_0
- pool_size = 2x2
- Conv2D: bnr_complex_conv2d_1
- filters = 32
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- BatchNormalization: bnr_complex_batch_normalization
- Activation: bnr_complex_activation
- activation = relu (rectified linear unit)
- MaxPooling2D: bnr_complex_max_pooling2d_1
- pool_size = 2x2
- Conv2D: bnr_complex_conv2d_2
- filters = 64
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- MaxPooling2D: bnr_complex_max_pooling2d_2
- pool_size = 2x2
- Flatten: bnr_complex_flatten
- Dense: bnr_complex_dense_0
- units = 128
- activation = relu (rectified linear unit)
- Dense: bnr_complex_dense_1
- units = 4
- activation = softmax
- Conv2D: bnr_complex_conv2d_0
- Additional fixed hyperparameters used during model compilation are as follows:
- loss = categorical_crossentropy
- optimizer = adam (adaptive moment estimation)
- metrics = recall
In [116]:
##################################
# Formulating the network architecture
# for a simple CNN with batch normalization regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_bnr_simple = Sequential(name="model_bnr_simple")
model_bnr_simple.add(Conv2D(filters=8, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="bnr_simple_conv2d_0"))
model_bnr_simple.add(MaxPooling2D(pool_size=(2, 2), name="bnr_simple_max_pooling2d_0"))
model_bnr_simple.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', name="bnr_simple_conv2d_1"))
model_bnr_simple.add(BatchNormalization(name="bnr_simple_batch_normalization"))
model_bnr_simple.add(Activation('relu', name="bnr_simple_activation"))
model_bnr_simple.add(MaxPooling2D(pool_size=(2, 2), name="bnr_simple_max_pooling2d_1"))
model_bnr_simple.add(Flatten(name="bnr_simple_flatten"))
model_bnr_simple.add(Dense(units=32, activation='relu', name="bnr_simple_dense_0"))
model_bnr_simple.add(Dense(units=num_classes, activation='softmax', name="bnr_simple_dense_1"))
In [117]:
##################################
# Displaying the model summary
# for a simple CNN with batch normalization regularization
##################################
print(model_bnr_simple.summary())
Model: "model_bnr_simple"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ bnr_simple_conv2d_0 (Conv2D) │ (None, 227, 227, 8) │ 80 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_simple_max_pooling2d_0 │ (None, 113, 113, 8) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_simple_conv2d_1 (Conv2D) │ (None, 113, 113, 16) │ 1,168 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_simple_batch_normalization │ (None, 113, 113, 16) │ 64 │ │ (BatchNormalization) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_simple_activation (Activation) │ (None, 113, 113, 16) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_simple_max_pooling2d_1 │ (None, 56, 56, 16) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_simple_flatten (Flatten) │ (None, 50176) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_simple_dense_0 (Dense) │ (None, 32) │ 1,605,664 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_simple_dense_1 (Dense) │ (None, 4) │ 132 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 1,607,108 (6.13 MB)
Trainable params: 1,607,076 (6.13 MB)
Non-trainable params: 32 (128.00 B)
None
In [118]:
##################################
# Displaying the model layers
# for a simple CNN with batch normalization regularization
##################################
model_bnr_simple_layer_names = [layer.name for layer in model_bnr_simple.layers]
print("Layer Names:", model_bnr_simple_layer_names)
Layer Names: ['bnr_simple_conv2d_0', 'bnr_simple_max_pooling2d_0', 'bnr_simple_conv2d_1', 'bnr_simple_batch_normalization', 'bnr_simple_activation', 'bnr_simple_max_pooling2d_1', 'bnr_simple_flatten', 'bnr_simple_dense_0', 'bnr_simple_dense_1']
In [119]:
##################################
# Displaying the number of weights
# for each model layer
# for a simple CNN with batch normalization regularization
##################################
for layer in model_bnr_simple.layers:
if hasattr(layer, 'weights'):
print(f"Layer: {layer.name}, Number of Weights: {len(layer.get_weights())}")
Layer: bnr_simple_conv2d_0, Number of Weights: 2 Layer: bnr_simple_max_pooling2d_0, Number of Weights: 0 Layer: bnr_simple_conv2d_1, Number of Weights: 2 Layer: bnr_simple_batch_normalization, Number of Weights: 4 Layer: bnr_simple_activation, Number of Weights: 0 Layer: bnr_simple_max_pooling2d_1, Number of Weights: 0 Layer: bnr_simple_flatten, Number of Weights: 0 Layer: bnr_simple_dense_0, Number of Weights: 2 Layer: bnr_simple_dense_1, Number of Weights: 2
In [120]:
##################################
# Displaying the number of weights
# for each model layer
# for a simple CNN with batch normalization regularization
##################################
total_parameters = 0
for layer in model_bnr_simple.layers:
layer_parameters = layer.count_params()
total_parameters += layer_parameters
print(f"Layer: {layer.name}, Parameters: {layer_parameters}")
print("\nTotal Parameters in the Model:", total_parameters)
Layer: bnr_simple_conv2d_0, Parameters: 80 Layer: bnr_simple_max_pooling2d_0, Parameters: 0 Layer: bnr_simple_conv2d_1, Parameters: 1168 Layer: bnr_simple_batch_normalization, Parameters: 64 Layer: bnr_simple_activation, Parameters: 0 Layer: bnr_simple_max_pooling2d_1, Parameters: 0 Layer: bnr_simple_flatten, Parameters: 0 Layer: bnr_simple_dense_0, Parameters: 1605664 Layer: bnr_simple_dense_1, Parameters: 132 Total Parameters in the Model: 1607108
In [121]:
##################################
# Formulating the network architecture
# for a complex CNN with batch normalization regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_bnr_complex = Sequential(name="model_bnr_complex")
model_bnr_complex.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="bnr_complex_conv2d_0"))
model_bnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="bnr_complex_max_pooling2d_0"))
model_bnr_complex.add(Conv2D(filters=32, kernel_size=(3, 3), padding = 'Same', activation='relu', name="bnr_complex_conv2d_1"))
model_bnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="bnr_complex_max_pooling2d_1"))
model_bnr_complex.add(Conv2D(filters=64, kernel_size=(3, 3), padding = 'Same', activation='relu', name="bnr_complex_conv2d_2"))
model_bnr_complex.add(BatchNormalization(name="bnr_complex_batch_normalization"))
model_bnr_complex.add(Activation('relu', name="bnr_complex_activation"))
model_bnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="bnr_complex_max_pooling2d_2"))
model_bnr_complex.add(Flatten(name="bnr_complex_flatten"))
model_bnr_complex.add(Dense(units=128, activation='relu', name="bnr_complex_dense_0"))
model_bnr_complex.add(Dense(units=num_classes, activation='softmax', name="bnr_complex_dense_1"))
In [122]:
##################################
# Displaying the model summary
# for a complex CNN with batch normalization regularization
##################################
print(model_bnr_complex.summary())
Model: "model_bnr_complex"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ bnr_complex_conv2d_0 (Conv2D) │ (None, 227, 227, 16) │ 160 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_max_pooling2d_0 │ (None, 113, 113, 16) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_conv2d_1 (Conv2D) │ (None, 113, 113, 32) │ 4,640 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_max_pooling2d_1 │ (None, 56, 56, 32) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_conv2d_2 (Conv2D) │ (None, 56, 56, 64) │ 18,496 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_batch_normalization │ (None, 56, 56, 64) │ 256 │ │ (BatchNormalization) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_activation (Activation) │ (None, 56, 56, 64) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_max_pooling2d_2 │ (None, 28, 28, 64) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_flatten (Flatten) │ (None, 50176) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_dense_0 (Dense) │ (None, 128) │ 6,422,656 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ bnr_complex_dense_1 (Dense) │ (None, 4) │ 516 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 6,446,724 (24.59 MB)
Trainable params: 6,446,596 (24.59 MB)
Non-trainable params: 128 (512.00 B)
None
In [123]:
##################################
# Displaying the model layers
# for a complex CNN with batch normalization regularization
##################################
model_bnr_complex_layer_names = [layer.name for layer in model_bnr_complex.layers]
print("Layer Names:", model_bnr_complex_layer_names)
Layer Names: ['bnr_complex_conv2d_0', 'bnr_complex_max_pooling2d_0', 'bnr_complex_conv2d_1', 'bnr_complex_max_pooling2d_1', 'bnr_complex_conv2d_2', 'bnr_complex_batch_normalization', 'bnr_complex_activation', 'bnr_complex_max_pooling2d_2', 'bnr_complex_flatten', 'bnr_complex_dense_0', 'bnr_complex_dense_1']
In [124]:
##################################
# Displaying the number of weights
# for each model layer
# for a complex CNN with batch normalization regularization
##################################
for layer in model_bnr_complex.layers:
if hasattr(layer, 'weights'):
print(f"Layer: {layer.name}, Number of Weights: {len(layer.get_weights())}")
Layer: bnr_complex_conv2d_0, Number of Weights: 2 Layer: bnr_complex_max_pooling2d_0, Number of Weights: 0 Layer: bnr_complex_conv2d_1, Number of Weights: 2 Layer: bnr_complex_max_pooling2d_1, Number of Weights: 0 Layer: bnr_complex_conv2d_2, Number of Weights: 2 Layer: bnr_complex_batch_normalization, Number of Weights: 4 Layer: bnr_complex_activation, Number of Weights: 0 Layer: bnr_complex_max_pooling2d_2, Number of Weights: 0 Layer: bnr_complex_flatten, Number of Weights: 0 Layer: bnr_complex_dense_0, Number of Weights: 2 Layer: bnr_complex_dense_1, Number of Weights: 2
In [125]:
##################################
# Displaying the number of weights
# for each model layer
# for a complex CNN with batch normalization regularization
##################################
total_parameters = 0
for layer in model_bnr_complex.layers:
layer_parameters = layer.count_params()
total_parameters += layer_parameters
print(f"Layer: {layer.name}, Parameters: {layer_parameters}")
print("\nTotal Parameters in the Model:", total_parameters)
Layer: bnr_complex_conv2d_0, Parameters: 160 Layer: bnr_complex_max_pooling2d_0, Parameters: 0 Layer: bnr_complex_conv2d_1, Parameters: 4640 Layer: bnr_complex_max_pooling2d_1, Parameters: 0 Layer: bnr_complex_conv2d_2, Parameters: 18496 Layer: bnr_complex_batch_normalization, Parameters: 256 Layer: bnr_complex_activation, Parameters: 0 Layer: bnr_complex_max_pooling2d_2, Parameters: 0 Layer: bnr_complex_flatten, Parameters: 0 Layer: bnr_complex_dense_0, Parameters: 6422656 Layer: bnr_complex_dense_1, Parameters: 516 Total Parameters in the Model: 6446724
1.6.2.4 CNN With Dropout and Batch Normalization Regularization ¶
- The simple model contains 10 layers with fixed hyperparameters as follows:
- Conv2D: cdrbnr_simple_conv2d_0
- filters = 8
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- input_shape = 227x227x1
- MaxPooling2D: cdrbnr_simple_max_pooling2d_0
- pool_size = 2x2
- Conv2D: cdrbnr_simple_conv2d_1
- filters = 16
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- BatchNormalization: cdrbnr_simple_batch_normalization
- Activation: cdrbnr_simple_activation
- activation = relu (rectified linear unit)
- MaxPooling2D: cdrbnr_simple_max_pooling2d_1
- pool_size = 2x2
- Flatten: cdrbnr_simple_flatten
- Dense: cdrbnr_simple_dense_0
- units = 32
- activation = relu (rectified linear unit)
- Dropout: cdrbnr_simple_dropout
- rate = 0.25
- Dense: cdrbnr_simple_dense_1
- units = 4
- activation = softmax
- Conv2D: cdrbnr_simple_conv2d_0
- The complex model contains 11 layers with fixed hyperparameters as follows:
- Conv2D: cdrbnr_complex_conv2d_0
- filters = 16
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- input_shape = 227x227x1
- MaxPooling2D: cdrbnr_complex_max_pooling2d_0
- pool_size = 2x2
- Conv2D: cdrbnr_complex_conv2d_1
- filters = 32
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- BatchNormalization: cdrbnr_complex_batch_normalization
- Activation: cdrbnr_complex_activation
- activation = relu (rectified linear unit)
- MaxPooling2D: cdrbnr_complex_max_pooling2d_1
- pool_size = 2x2
- Conv2D: cdrbnr_complex_conv2d_2
- filters = 64
- kernel_size = 3x3
- activation = relu (rectified linear unit)
- padding = same (output size equals input size)
- MaxPooling2D: cdrbnr_complex_max_pooling2d_2
- pool_size = 2x2
- Flatten: cdrbnr_complex_flatten
- Dense: cdrbnr_complex_dense_0
- units = 128
- activation = relu (rectified linear unit)
- Dropout: cdrbnr_complex_dropout
- rate = 0.25
- Dense: cdrbnr_complex_dense_1
- units = 4
- activation = softmax
- Conv2D: cdrbnr_complex_conv2d_0
- Additional fixed hyperparameters used during model compilation are as follows:
- loss = categorical_crossentropy
- optimizer = adam (adaptive moment estimation)
- metrics = recall
In [126]:
##################################
# Formulating the network architecture
# for a simple CNN with dropout and batch normalization regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_cdrbnr_simple = Sequential(name="model_cdrbnr_simple")
model_cdrbnr_simple.add(Conv2D(filters=8, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="cdrbnr_simple_conv2d_0"))
model_cdrbnr_simple.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_simple_max_pooling2d_0"))
model_cdrbnr_simple.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', name="cdrbnr_simple_conv2d_1"))
model_cdrbnr_simple.add(BatchNormalization(name="cdrbnr_simple_batch_normalization"))
model_cdrbnr_simple.add(Activation('relu', name="cdrbnr_simple_activation"))
model_cdrbnr_simple.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_simple_max_pooling2d_1"))
model_cdrbnr_simple.add(Flatten(name="cdrbnr_simple_flatten"))
model_cdrbnr_simple.add(Dense(units=32, activation='relu', name="cdrbnr_simple_dense_0"))
model_cdrbnr_simple.add(Dropout(rate=0.30, name="cdrbnr_simple_dropout"))
model_cdrbnr_simple.add(Dense(units=num_classes, activation='softmax', name="cdrbnr_simple_dense_1"))
In [127]:
##################################
# Displaying the model summary
# for a simple CNN with dropout and batch normalization regularization
##################################
print(model_cdrbnr_simple.summary())
Model: "model_cdrbnr_simple"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ cdrbnr_simple_conv2d_0 (Conv2D) │ (None, 227, 227, 8) │ 80 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_simple_max_pooling2d_0 │ (None, 113, 113, 8) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_simple_conv2d_1 (Conv2D) │ (None, 113, 113, 16) │ 1,168 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_simple_batch_normalization │ (None, 113, 113, 16) │ 64 │ │ (BatchNormalization) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_simple_activation │ (None, 113, 113, 16) │ 0 │ │ (Activation) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_simple_max_pooling2d_1 │ (None, 56, 56, 16) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_simple_flatten (Flatten) │ (None, 50176) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_simple_dense_0 (Dense) │ (None, 32) │ 1,605,664 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_simple_dropout (Dropout) │ (None, 32) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_simple_dense_1 (Dense) │ (None, 4) │ 132 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 1,607,108 (6.13 MB)
Trainable params: 1,607,076 (6.13 MB)
Non-trainable params: 32 (128.00 B)
None
In [128]:
##################################
# Displaying the model layers
# for a simple CNN with dropout and batch normalization regularization
##################################
model_cdrbnr_simple_layer_names = [layer.name for layer in model_cdrbnr_simple.layers]
print("Layer Names:", model_cdrbnr_simple_layer_names)
Layer Names: ['cdrbnr_simple_conv2d_0', 'cdrbnr_simple_max_pooling2d_0', 'cdrbnr_simple_conv2d_1', 'cdrbnr_simple_batch_normalization', 'cdrbnr_simple_activation', 'cdrbnr_simple_max_pooling2d_1', 'cdrbnr_simple_flatten', 'cdrbnr_simple_dense_0', 'cdrbnr_simple_dropout', 'cdrbnr_simple_dense_1']
In [129]:
##################################
# Displaying the number of weights
# for each model layer
# for a simple CNN with dropout and batch normalization regularization
##################################
for layer in model_cdrbnr_simple.layers:
if hasattr(layer, 'weights'):
print(f"Layer: {layer.name}, Number of Weights: {len(layer.get_weights())}")
Layer: cdrbnr_simple_conv2d_0, Number of Weights: 2 Layer: cdrbnr_simple_max_pooling2d_0, Number of Weights: 0 Layer: cdrbnr_simple_conv2d_1, Number of Weights: 2 Layer: cdrbnr_simple_batch_normalization, Number of Weights: 4 Layer: cdrbnr_simple_activation, Number of Weights: 0 Layer: cdrbnr_simple_max_pooling2d_1, Number of Weights: 0 Layer: cdrbnr_simple_flatten, Number of Weights: 0 Layer: cdrbnr_simple_dense_0, Number of Weights: 2 Layer: cdrbnr_simple_dropout, Number of Weights: 0 Layer: cdrbnr_simple_dense_1, Number of Weights: 2
In [130]:
##################################
# Displaying the number of weights
# for each model layer
# for a simple CNN with dropout and batch normalization regularization
##################################
total_parameters = 0
for layer in model_cdrbnr_simple.layers:
layer_parameters = layer.count_params()
total_parameters += layer_parameters
print(f"Layer: {layer.name}, Parameters: {layer_parameters}")
print("\nTotal Parameters in the Model:", total_parameters)
Layer: cdrbnr_simple_conv2d_0, Parameters: 80 Layer: cdrbnr_simple_max_pooling2d_0, Parameters: 0 Layer: cdrbnr_simple_conv2d_1, Parameters: 1168 Layer: cdrbnr_simple_batch_normalization, Parameters: 64 Layer: cdrbnr_simple_activation, Parameters: 0 Layer: cdrbnr_simple_max_pooling2d_1, Parameters: 0 Layer: cdrbnr_simple_flatten, Parameters: 0 Layer: cdrbnr_simple_dense_0, Parameters: 1605664 Layer: cdrbnr_simple_dropout, Parameters: 0 Layer: cdrbnr_simple_dense_1, Parameters: 132 Total Parameters in the Model: 1607108
In [131]:
##################################
# Formulating the network architecture
# for a complex CNN with dropout and batch normalization regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_cdrbnr_complex = Sequential(name="model_cdrbnr_complex")
model_cdrbnr_complex.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="cdrbnr_complex_conv2d_0"))
model_cdrbnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_complex_max_pooling2d_0"))
model_cdrbnr_complex.add(Conv2D(filters=32, kernel_size=(3, 3), padding = 'Same', activation='relu', name="cdrbnr_complex_conv2d_1"))
model_cdrbnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_complex_max_pooling2d_1"))
model_cdrbnr_complex.add(Conv2D(filters=64, kernel_size=(3, 3), padding = 'Same', activation='relu', name="cdrbnr_complex_conv2d_2"))
model_cdrbnr_complex.add(BatchNormalization(name="cdrbnr_complex_batch_normalization"))
model_cdrbnr_complex.add(Activation('relu', name="cdrbnr_complex_activation"))
model_cdrbnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_complex_max_pooling2d_2"))
model_cdrbnr_complex.add(Flatten(name="cdrbnr_complex_flatten"))
model_cdrbnr_complex.add(Dense(units=128, activation='relu', name="cdrbnr_complex_dense_0"))
model_cdrbnr_complex.add(Dropout(rate=0.30, name="cdrbnr_complex_dropout"))
model_cdrbnr_complex.add(Dense(units=num_classes, activation='softmax', name="cdrbnr_complex_dense_1"))
In [132]:
##################################
# Displaying the model summary
# for a complex CNN with dropout and batch normalization regularization
##################################
print(model_cdrbnr_complex.summary())
Model: "model_cdrbnr_complex"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ cdrbnr_complex_conv2d_0 (Conv2D) │ (None, 227, 227, 16) │ 160 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_max_pooling2d_0 │ (None, 113, 113, 16) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_conv2d_1 (Conv2D) │ (None, 113, 113, 32) │ 4,640 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_max_pooling2d_1 │ (None, 56, 56, 32) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_conv2d_2 (Conv2D) │ (None, 56, 56, 64) │ 18,496 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_batch_normalization │ (None, 56, 56, 64) │ 256 │ │ (BatchNormalization) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_activation │ (None, 56, 56, 64) │ 0 │ │ (Activation) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_max_pooling2d_2 │ (None, 28, 28, 64) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_flatten (Flatten) │ (None, 50176) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_dense_0 (Dense) │ (None, 128) │ 6,422,656 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_dropout (Dropout) │ (None, 128) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_dense_1 (Dense) │ (None, 4) │ 516 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 6,446,724 (24.59 MB)
Trainable params: 6,446,596 (24.59 MB)
Non-trainable params: 128 (512.00 B)
None
In [133]:
##################################
# Displaying the model layers
# for a complex CNN with dropout and batch normalization regularization
##################################
model_cdrbnr_complex_layer_names = [layer.name for layer in model_cdrbnr_complex.layers]
print("Layer Names:", model_cdrbnr_complex_layer_names)
Layer Names: ['cdrbnr_complex_conv2d_0', 'cdrbnr_complex_max_pooling2d_0', 'cdrbnr_complex_conv2d_1', 'cdrbnr_complex_max_pooling2d_1', 'cdrbnr_complex_conv2d_2', 'cdrbnr_complex_batch_normalization', 'cdrbnr_complex_activation', 'cdrbnr_complex_max_pooling2d_2', 'cdrbnr_complex_flatten', 'cdrbnr_complex_dense_0', 'cdrbnr_complex_dropout', 'cdrbnr_complex_dense_1']
In [134]:
##################################
# Displaying the number of weights
# for each model layer
# for a complex CNN with dropout and batch normalization regularization
##################################
for layer in model_cdrbnr_complex.layers:
if hasattr(layer, 'weights'):
print(f"Layer: {layer.name}, Number of Weights: {len(layer.get_weights())}")
Layer: cdrbnr_complex_conv2d_0, Number of Weights: 2 Layer: cdrbnr_complex_max_pooling2d_0, Number of Weights: 0 Layer: cdrbnr_complex_conv2d_1, Number of Weights: 2 Layer: cdrbnr_complex_max_pooling2d_1, Number of Weights: 0 Layer: cdrbnr_complex_conv2d_2, Number of Weights: 2 Layer: cdrbnr_complex_batch_normalization, Number of Weights: 4 Layer: cdrbnr_complex_activation, Number of Weights: 0 Layer: cdrbnr_complex_max_pooling2d_2, Number of Weights: 0 Layer: cdrbnr_complex_flatten, Number of Weights: 0 Layer: cdrbnr_complex_dense_0, Number of Weights: 2 Layer: cdrbnr_complex_dropout, Number of Weights: 0 Layer: cdrbnr_complex_dense_1, Number of Weights: 2
In [135]:
##################################
# Displaying the number of weights
# for each model layer
# for a complex CNN with dropout and batch normalization regularization
##################################
total_parameters = 0
for layer in model_cdrbnr_complex.layers:
layer_parameters = layer.count_params()
total_parameters += layer_parameters
print(f"Layer: {layer.name}, Parameters: {layer_parameters}")
print("\nTotal Parameters in the Model:", total_parameters)
Layer: cdrbnr_complex_conv2d_0, Parameters: 160 Layer: cdrbnr_complex_max_pooling2d_0, Parameters: 0 Layer: cdrbnr_complex_conv2d_1, Parameters: 4640 Layer: cdrbnr_complex_max_pooling2d_1, Parameters: 0 Layer: cdrbnr_complex_conv2d_2, Parameters: 18496 Layer: cdrbnr_complex_batch_normalization, Parameters: 256 Layer: cdrbnr_complex_activation, Parameters: 0 Layer: cdrbnr_complex_max_pooling2d_2, Parameters: 0 Layer: cdrbnr_complex_flatten, Parameters: 0 Layer: cdrbnr_complex_dense_0, Parameters: 6422656 Layer: cdrbnr_complex_dropout, Parameters: 0 Layer: cdrbnr_complex_dense_1, Parameters: 516 Total Parameters in the Model: 6446724
1.6.3 CNN With No Regularization Model Fitting | Hyperparameter Tuning | Validation ¶
- The simple model contained 1,607,044 trainable parameters broken down per layer as follows:
- Conv2D: nr_simple_conv2d_0
- output size = 227x227x8
- number of parameters = 80
- MaxPooling2D: nr_simple_max_pooling2d_0
- output size = 113x113x8
- number of parameters = 0
- Conv2D: nr_simple_conv2d_1
- output size = 113x113x16
- number of parameters = 1,168
- MaxPooling2D: nr_simple_max_pooling2d_1
- output size = 56x56x16
- number of parameters = 0
- Flatten: nr_simple_flatten
- output size = 50,176
- number of parameters = 0
- Dense: nr_simple_dense_0
- output size = 32
- number of parameters = 1,605,664
- Dense: nr_simple_dense_1
- output size = 4
- number of parameters = 132
- Conv2D: nr_simple_conv2d_0
- The complex model contained 6,446,468 trainable parameters broken down per layer as follows:
- Conv2D: nr_complex_conv2d_0
- output size = 227x227x16
- number of parameters = 160
- MaxPooling2D: nr_complex_max_pooling2d_0
- output size = 113x113x16
- number of parameters = 0
- Conv2D: nr_complex_conv2d_1
- output size = 113x113x32
- number of parameters = 4,640
- MaxPooling2D: nr_complex_max_pooling2d_1
- output size = 56x56x32
- number of parameters = 0
- Conv2D: nr_complex_conv2d_2
- output size = 56x56x64
- number of parameters = 18,496
- MaxPooling2D: nr_complex_max_pooling2d_2
- output size = 28x28x64
- number of parameters = 0
- Flatten: nr_complex_flatten
- output size = 50,176
- number of parameters = 0
- Dense: nr_complex_dense_0
- output size = 128
- number of parameters = 6,422,656
- Dense: nr_complex_dense_1
- output size = 4
- number of parameters = 516
- Conv2D: nr_complex_conv2d_0
- The model performance on the validation set for all image categories is summarized as follows:
- Simple
- Precision = 0.8047
- Recall = 0.7926
- F1 Score = 0.7946
- Complex
- Precision = 0.7963
- Recall = 0.7960
- F1 Score = 0.7944
- Simple
In [136]:
##################################
# Formulating the network architecture
# for a simple CNN with no regularization
##################################
set_seed()
batch_size = 32
model_nr_simple = Sequential(name="model_nr_simple")
model_nr_simple.add(Conv2D(filters=8, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="nr_simple_conv2d_0"))
model_nr_simple.add(MaxPooling2D(pool_size=(2, 2), name="nr_simple_max_pooling2d_0"))
model_nr_simple.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', name="nr_simple_conv2d_1"))
model_nr_simple.add(MaxPooling2D(pool_size=(2, 2), name="nr_simple_max_pooling2d_1"))
model_nr_simple.add(Flatten(name="nr_simple_flatten"))
model_nr_simple.add(Dense(units=32, activation='relu', name="nr_simple_dense_0"))
model_nr_simple.add(Dense(units=num_classes, activation='softmax', name="nr_simple_dense_1"))
##################################
# Compiling the network layers
##################################
model_nr_simple.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [137]:
##################################
# Fitting the model
# for a simple CNN with no regularization
##################################
epochs = 20
set_seed()
model_nr_simple_history = model_nr_simple.fit(train_gen,
steps_per_epoch=len(train_gen)+1,
validation_steps=len(val_gen)+1,
validation_data=val_gen,
epochs=epochs,
verbose=1,
callbacks=[early_stopping, reduce_lr, nr_simple_model_checkpoint])
Epoch 1/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 37s 245ms/step - loss: 0.8697 - recall: 0.4612 - val_loss: 0.9379 - val_recall: 0.6556 - learning_rate: 0.0010 Epoch 2/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 42s 252ms/step - loss: 0.4279 - recall: 0.8129 - val_loss: 0.8684 - val_recall: 0.6792 - learning_rate: 0.0010 Epoch 3/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 40s 246ms/step - loss: 0.3339 - recall: 0.8637 - val_loss: 0.8071 - val_recall: 0.7239 - learning_rate: 0.0010 Epoch 4/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 40s 240ms/step - loss: 0.3062 - recall: 0.8771 - val_loss: 0.9367 - val_recall: 0.7528 - learning_rate: 0.0010 Epoch 5/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 236ms/step - loss: 0.2505 - recall: 0.9024 - val_loss: 0.8099 - val_recall: 0.7450 - learning_rate: 0.0010 Epoch 6/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 42s 244ms/step - loss: 0.2282 - recall: 0.9033 - val_loss: 0.7319 - val_recall: 0.7862 - learning_rate: 0.0010 Epoch 7/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 42s 249ms/step - loss: 0.1857 - recall: 0.9301 - val_loss: 0.8285 - val_recall: 0.7783 - learning_rate: 0.0010 Epoch 8/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 35s 242ms/step - loss: 0.1783 - recall: 0.9361 - val_loss: 0.8437 - val_recall: 0.7642 - learning_rate: 0.0010 Epoch 9/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 237ms/step - loss: 0.1366 - recall: 0.9491 - val_loss: 0.8675 - val_recall: 0.8089 - learning_rate: 0.0010 Epoch 10/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 237ms/step - loss: 0.1127 - recall: 0.9611 - val_loss: 0.7600 - val_recall: 0.8186 - learning_rate: 1.0000e-04 Epoch 11/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 35s 241ms/step - loss: 0.0880 - recall: 0.9663 - val_loss: 0.7769 - val_recall: 0.8177 - learning_rate: 1.0000e-04 Epoch 12/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 107s 748ms/step - loss: 0.1055 - recall: 0.9612 - val_loss: 0.7722 - val_recall: 0.8221 - learning_rate: 1.0000e-04 Epoch 13/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 70s 244ms/step - loss: 0.0787 - recall: 0.9733 - val_loss: 0.7732 - val_recall: 0.8221 - learning_rate: 1.0000e-05 Epoch 14/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 35s 245ms/step - loss: 0.0926 - recall: 0.9680 - val_loss: 0.7768 - val_recall: 0.8221 - learning_rate: 1.0000e-05 Epoch 15/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 239ms/step - loss: 0.0824 - recall: 0.9691 - val_loss: 0.7803 - val_recall: 0.8212 - learning_rate: 1.0000e-05 Epoch 16/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 33s 233ms/step - loss: 0.0939 - recall: 0.9701 - val_loss: 0.7808 - val_recall: 0.8203 - learning_rate: 1.0000e-06
In [138]:
##################################
# Evaluating the model
# for a simple CNN with no regularization
# on the independent validation set
##################################
model_nr_simple_y_pred_val = model_nr_simple.predict(val_gen)
36/36 ━━━━━━━━━━━━━━━━━━━━ 4s 110ms/step
In [139]:
##################################
# Plotting the loss profile
# for a simple CNN with no regularization
# on the training and validation sets
##################################
plot_training_history(model_nr_simple_history, 'Simple CNN With No Regularization : ')
In [140]:
##################################
# Consolidating the predictions
# for a simple CNN with no regularization
# on the validation set
##################################
model_nr_simple_predictions_val = np.array(list(map(lambda x: np.argmax(x), model_nr_simple_y_pred_val)))
model_nr_simple_y_true_val = val_gen.classes
##################################
# Formulating the confusion matrix
# for a simple CNN with no regularization
# on the validation set
##################################
cmatrix_val = pd.DataFrame(confusion_matrix(model_nr_simple_y_true_val, model_nr_simple_predictions_val), columns=classes, index =classes)
##################################
# Plotting the confusion matrix
# for a simple CNN with no regularization
# on the validation set
##################################
plt.figure(figsize=(10, 6))
ax = sns.heatmap(cmatrix_val, annot = True, fmt = 'g' ,vmin = 0, vmax = 250,cmap = 'icefire')
ax.set_xlabel('Predicted',fontsize = 14,weight = 'bold')
ax.set_xticklabels(ax.get_xticklabels(),rotation =0)
ax.set_ylabel('Actual',fontsize = 14,weight = 'bold')
ax.set_yticklabels(ax.get_yticklabels(),rotation =0)
ax.set_title('Simple CNN With No Regularization : Validation Set Confusion Matrix',fontsize = 14, weight = 'bold',pad=20);
##################################
# Resetting all states generated by Keras
##################################
keras.backend.clear_session()
WARNING:tensorflow:From D:\Github_Codes\ProjectPortfolio\Portfolio_Project_56\mdeploy_venv\Lib\site-packages\keras\src\backend\common\global_state.py:82: The name tf.reset_default_graph is deprecated. Please use tf.compat.v1.reset_default_graph instead.
In [141]:
##################################
# Calculating the model accuracy
# for a simple CNN with no regularization
# for the entire validation set
##################################
model_nr_simple_acc_val = accuracy_score(model_nr_simple_y_true_val, model_nr_simple_predictions_val)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with no regularization
# for the entire validation set
##################################
model_nr_simple_results_all_val = precision_recall_fscore_support(model_nr_simple_y_true_val, model_nr_simple_predictions_val, average='macro',zero_division = 1)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with no regularization
# for each category of the validation set
##################################
model_nr_simple_results_class_val = precision_recall_fscore_support(model_nr_simple_y_true_val, model_nr_simple_predictions_val, average=None, zero_division = 1)
##################################
# Consolidating all model evaluation metrics
# for a simple CNN with no regularization
##################################
metric_columns = ['Precision','Recall','F-Score','Support']
model_nr_simple_all_df_val = pd.concat([pd.DataFrame(list(model_nr_simple_results_class_val)).T,pd.DataFrame(list(model_nr_simple_results_all_val)).T])
model_nr_simple_all_df_val.columns = metric_columns
model_nr_simple_all_df_val.index = ['No Tumor', 'Glioma', 'Meningioma', 'Pituitary', 'Total']
print('Simple CNN With No Regularization : Validation Set Classification Performance')
model_nr_simple_all_df_val
Simple CNN With No Regularization : Validation Set Classification Performance
Out[141]:
| Precision | Recall | F-Score | Support | |
|---|---|---|---|---|
| No Tumor | 0.893238 | 0.786834 | 0.836667 | 319.0 |
| Glioma | 0.928571 | 0.787879 | 0.852459 | 264.0 |
| Meningioma | 0.624573 | 0.685393 | 0.653571 | 267.0 |
| Pituitary | 0.772595 | 0.910653 | 0.835962 | 291.0 |
| Total | 0.804744 | 0.792690 | 0.794665 | NaN |
In [142]:
##################################
# Consolidating all model evaluation metrics
# for a simple CNN with no regularization
##################################
model_nr_simple_model_list_val = []
model_nr_simple_measure_list_val = []
model_nr_simple_category_list_val = []
model_nr_simple_value_list_val = []
model_nr_simple_dataset_list_val = []
for i in range(3):
for j in range(5):
model_nr_simple_model_list_val.append('CNN_NR_Simple')
model_nr_simple_measure_list_val.append(metric_columns[i])
model_nr_simple_category_list_val.append(model_nr_simple_all_df_val.index[j])
model_nr_simple_value_list_val.append(model_nr_simple_all_df_val.iloc[j,i])
model_nr_simple_dataset_list_val.append('Validation')
model_nr_simple_all_summary_val = pd.DataFrame(zip(model_nr_simple_model_list_val,
model_nr_simple_measure_list_val,
model_nr_simple_category_list_val,
model_nr_simple_value_list_val,
model_nr_simple_dataset_list_val),
columns=['CNN.Model.Name',
'Model.Metric',
'Image.Category',
'Metric.Value',
'Data.Set'])
In [143]:
##################################
# Formulating the network architecture
# for a complex CNN with no regularization
##################################
set_seed()
batch_size = 32
model_nr_complex = Sequential(name="model_nr_complex")
model_nr_complex.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="nr_complex_conv2d_0"))
model_nr_complex.add(MaxPooling2D(pool_size=(2, 2), name="nr_complex_max_pooling2d_0"))
model_nr_complex.add(Conv2D(filters=32, kernel_size=(3, 3), padding = 'Same', activation='relu', name="nr_complex_conv2d_1"))
model_nr_complex.add(MaxPooling2D(pool_size=(2, 2), name="nr_complex_max_pooling2d_1"))
model_nr_complex.add(Conv2D(filters=64, kernel_size=(3, 3), padding = 'Same', activation='relu', name="nr_complex_conv2d_2"))
model_nr_complex.add(MaxPooling2D(pool_size=(2, 2), name="nr_complex_max_pooling2d_2"))
model_nr_complex.add(Flatten(name="nr_complex_flatten"))
model_nr_complex.add(Dense(units=128, activation='relu', name="nr_complex_dense_0"))
model_nr_complex.add(Dense(units=num_classes, activation='softmax', name="nr_complex_dense_1"))
##################################
# Compiling the network layers
##################################
model_nr_complex.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [144]:
##################################
# Fitting the model
# for a complex CNN with no regularization
##################################
epochs = 20
set_seed()
model_nr_complex_history = model_nr_complex.fit(train_gen,
steps_per_epoch=len(train_gen)+1,
validation_steps=len(val_gen)+1,
validation_data=val_gen,
epochs=epochs,
verbose=1,
callbacks=[early_stopping, reduce_lr, nr_complex_model_checkpoint])
Epoch 1/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 373ms/step - loss: 1.0913 - recall: 0.3645 - val_loss: 0.8411 - val_recall: 0.6915 - learning_rate: 0.0010 Epoch 2/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 80s 357ms/step - loss: 0.4091 - recall: 0.8322 - val_loss: 0.8689 - val_recall: 0.6862 - learning_rate: 0.0010 Epoch 3/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 362ms/step - loss: 0.2674 - recall: 0.8948 - val_loss: 0.8096 - val_recall: 0.7327 - learning_rate: 0.0010 Epoch 4/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 82s 364ms/step - loss: 0.2156 - recall: 0.9202 - val_loss: 0.8086 - val_recall: 0.7862 - learning_rate: 0.0010 Epoch 5/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 82s 360ms/step - loss: 0.1748 - recall: 0.9339 - val_loss: 0.8040 - val_recall: 0.7625 - learning_rate: 0.0010 Epoch 6/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 83s 370ms/step - loss: 0.1469 - recall: 0.9431 - val_loss: 0.7236 - val_recall: 0.7984 - learning_rate: 0.0010 Epoch 7/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 80s 359ms/step - loss: 0.1025 - recall: 0.9621 - val_loss: 0.7801 - val_recall: 0.7993 - learning_rate: 0.0010 Epoch 8/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 360ms/step - loss: 0.0918 - recall: 0.9644 - val_loss: 0.9317 - val_recall: 0.8063 - learning_rate: 0.0010 Epoch 9/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 51s 357ms/step - loss: 0.0861 - recall: 0.9650 - val_loss: 0.8448 - val_recall: 0.8238 - learning_rate: 0.0010 Epoch 10/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 51s 357ms/step - loss: 0.0560 - recall: 0.9774 - val_loss: 0.8052 - val_recall: 0.8300 - learning_rate: 1.0000e-04 Epoch 11/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 362ms/step - loss: 0.0285 - recall: 0.9933 - val_loss: 0.8621 - val_recall: 0.8186 - learning_rate: 1.0000e-04 Epoch 12/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 53s 368ms/step - loss: 0.0294 - recall: 0.9919 - val_loss: 0.8798 - val_recall: 0.8256 - learning_rate: 1.0000e-04 Epoch 13/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 51s 353ms/step - loss: 0.0235 - recall: 0.9925 - val_loss: 0.8846 - val_recall: 0.8230 - learning_rate: 1.0000e-05 Epoch 14/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 51s 352ms/step - loss: 0.0297 - recall: 0.9903 - val_loss: 0.8888 - val_recall: 0.8247 - learning_rate: 1.0000e-05 Epoch 15/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 51s 353ms/step - loss: 0.0237 - recall: 0.9951 - val_loss: 0.9018 - val_recall: 0.8230 - learning_rate: 1.0000e-05 Epoch 16/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 51s 356ms/step - loss: 0.0283 - recall: 0.9913 - val_loss: 0.9021 - val_recall: 0.8238 - learning_rate: 1.0000e-06
In [145]:
##################################
# Evaluating the model
# for a complex CNN with no regularization
# on the independent validation set
##################################
model_nr_complex_y_pred_val = model_nr_complex.predict(val_gen)
36/36 ━━━━━━━━━━━━━━━━━━━━ 4s 120ms/step
In [146]:
##################################
# Plotting the loss profile
# for a complex CNN with no regularization
# on the training and validation sets
##################################
plot_training_history(model_nr_complex_history, 'Complex CNN With No Regularization : ')
In [147]:
##################################
# Consolidating the predictions
# for a complex CNN with no regularization
# on the validation set
##################################
model_nr_complex_predictions_val = np.array(list(map(lambda x: np.argmax(x), model_nr_complex_y_pred_val)))
model_nr_complex_y_true_val = val_gen.classes
##################################
# Formulating the confusion matrix
# for a complex CNN with no regularization
# on the validation set
##################################
cmatrix_val = pd.DataFrame(confusion_matrix(model_nr_complex_y_true_val, model_nr_complex_predictions_val), columns=classes, index =classes)
##################################
# Plotting the confusion matrix
# for a complex CNN with no regularization
# on the validation set
##################################
plt.figure(figsize=(10, 6))
ax = sns.heatmap(cmatrix_val, annot = True, fmt = 'g' ,vmin = 0, vmax = 250,cmap = 'icefire')
ax.set_xlabel('Predicted',fontsize = 14,weight = 'bold')
ax.set_xticklabels(ax.get_xticklabels(),rotation =0)
ax.set_ylabel('Actual',fontsize = 14,weight = 'bold')
ax.set_yticklabels(ax.get_yticklabels(),rotation =0)
ax.set_title('Complex CNN With No Regularization : Validation Set Confusion Matrix',fontsize = 14, weight = 'bold',pad=20);
##################################
# Resetting all states generated by Keras
##################################
keras.backend.clear_session()
In [148]:
##################################
# Calculating the model accuracy
# for a complex CNN with no regularization
# for the entire validation set
##################################
model_nr_complex_acc_val = accuracy_score(model_nr_complex_y_true_val, model_nr_complex_predictions_val)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with no regularization
# for the entire validation set
##################################
model_nr_complex_results_all_val = precision_recall_fscore_support(model_nr_complex_y_true_val, model_nr_complex_predictions_val, average='macro',zero_division = 1)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with no regularization
# for each category of the validation set
##################################
model_nr_complex_results_class_val = precision_recall_fscore_support(model_nr_complex_y_true_val, model_nr_complex_predictions_val, average=None, zero_division = 1)
##################################
# Consolidating all model evaluation metrics
# for a complex CNN with no regularization
##################################
metric_columns = ['Precision','Recall','F-Score','Support']
model_nr_complex_all_df_val = pd.concat([pd.DataFrame(list(model_nr_complex_results_class_val)).T,pd.DataFrame(list(model_nr_complex_results_all_val)).T])
model_nr_complex_all_df_val.columns = metric_columns
model_nr_complex_all_df_val.index = ['No Tumor', 'Glioma', 'Meningioma', 'Pituitary', 'Total']
print('Complex CNN With No Regularization : Validation Set Classification Performance')
model_nr_complex_all_df_val
Complex CNN With No Regularization : Validation Set Classification Performance
Out[148]:
| Precision | Recall | F-Score | Support | |
|---|---|---|---|---|
| No Tumor | 0.863057 | 0.849530 | 0.856240 | 319.0 |
| Glioma | 0.871486 | 0.821970 | 0.846004 | 264.0 |
| Meningioma | 0.655602 | 0.591760 | 0.622047 | 267.0 |
| Pituitary | 0.795252 | 0.920962 | 0.853503 | 291.0 |
| Total | 0.796349 | 0.796055 | 0.794449 | NaN |
In [149]:
##################################
# Consolidating all model evaluation metrics
# for a complex CNN with no regularization
##################################
model_nr_complex_model_list_val = []
model_nr_complex_measure_list_val = []
model_nr_complex_category_list_val = []
model_nr_complex_value_list_val = []
model_nr_complex_dataset_list_val = []
for i in range(3):
for j in range(5):
model_nr_complex_model_list_val.append('CNN_NR_Complex')
model_nr_complex_measure_list_val.append(metric_columns[i])
model_nr_complex_category_list_val.append(model_nr_complex_all_df_val.index[j])
model_nr_complex_value_list_val.append(model_nr_complex_all_df_val.iloc[j,i])
model_nr_complex_dataset_list_val.append('Validation')
model_nr_complex_all_summary_val = pd.DataFrame(zip(model_nr_complex_model_list_val,
model_nr_complex_measure_list_val,
model_nr_complex_category_list_val,
model_nr_complex_value_list_val,
model_nr_complex_dataset_list_val),
columns=['CNN.Model.Name',
'Model.Metric',
'Image.Category',
'Metric.Value',
'Data.Set'])
1.6.4 CNN With Dropout Regularization Model Fitting | Hyperparameter Tuning | Validation ¶
- The simple model contained 1,607,044 trainable parameters broken down per layer as follows:
- Conv2D: dr_simple_conv2d_0
- output size = 227x227x8
- number of parameters = 80
- MaxPooling2D: dr_simple_max_pooling2d_0
- output size = 113x113x8
- number of parameters = 0
- Conv2D: dr_simple_conv2d_1
- output size = 113x113x16
- number of parameters = 1,168
- MaxPooling2D: dr_simple_max_pooling2d_1
- output size = 56x56x16
- number of parameters = 0
- Flatten: dr_simple_flatten
- output size = 50,176
- number of parameters = 0
- Dense: dr_simple_dense_0
- output size = 32
- number of parameters = 1,605,664
- Dropout: dr_simple_dropout
- output size = 32
- number of parameters = 0
- Dense: dr_simple_dense_1
- output size = 4
- number of parameters = 132
- Conv2D: dr_simple_conv2d_0
- The complex model contained 6,446,468 trainable parameters broken down per layer as follows:
- Conv2D: dr_complex_conv2d_0
- output size = 227x227x16
- number of parameters = 160
- MaxPooling2D: dr_complex_max_pooling2d_0
- output size = 113x113x16
- number of parameters = 0
- Conv2D: dr_complex_conv2d_1
- output size = 113x113x32
- number of parameters = 4,640
- MaxPooling2D: dr_complex_max_pooling2d_1
- output size = 56x56x32
- number of parameters = 0
- Conv2D: dr_complex_conv2d_2
- output size = 56x56x64
- number of parameters = 18,496
- MaxPooling2D: dr_complex_max_pooling2d_2
- output size = 28x28x64
- number of parameters = 0
- Flatten: dr_complex_flatten
- output size = 50,176
- number of parameters = 0
- Dense: dr_complex_dense_0
- output size = 128
- number of parameters = 6,422,656
- Dropout: dr_complex_dropout
- output size = 128
- number of parameters = 0
- Dense: dr_complex_dense_1
- output size = 4
- number of parameters = 516
- Conv2D: dr_complex_conv2d_0
- The model performance on the validation set for all image categories is summarized as follows:
- Simple
- Precision = 0.7709
- Recall = 0.7576
- F1 Score = 0.7611
- Complex
- Precision = 0.7878
- Recall = 0.7897
- F1 Score = 0.7881
- Simple
In [150]:
##################################
# Formulating the network architecture
# for a simple CNN with dropout regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_dr_simple = Sequential(name="model_dr_simple")
model_dr_simple.add(Conv2D(filters=8, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="dr_simple_conv2d_0"))
model_dr_simple.add(MaxPooling2D(pool_size=(2, 2), name="dr_simple_max_pooling2d_0"))
model_dr_simple.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', name="dr_simple_conv2d_1"))
model_dr_simple.add(MaxPooling2D(pool_size=(2, 2), name="dr_simple_max_pooling2d_1"))
model_dr_simple.add(Flatten(name="dr_simple_flatten"))
model_dr_simple.add(Dense(units=32, activation='relu', name="dr_simple_dense_0"))
model_dr_simple.add(Dropout(rate=0.30, name="dr_simple_dropout"))
model_dr_simple.add(Dense(units=num_classes, activation='softmax', name="dr_simple_dense_1"))
##################################
# Compiling the network layers
##################################
model_dr_simple.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [151]:
##################################
# Fitting the model
# for a simple CNN with dropout regularization
##################################
epochs = 20
set_seed()
model_dr_simple_history = model_dr_simple.fit(train_gen,
steps_per_epoch=len(train_gen)+1,
validation_steps=len(val_gen)+1,
validation_data=val_gen,
epochs=epochs,
verbose=1,
callbacks=[early_stopping, reduce_lr, dr_simple_model_checkpoint])
Epoch 1/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 38s 253ms/step - loss: 1.3558 - recall: 0.1436 - val_loss: 1.0029 - val_recall: 0.4259 - learning_rate: 0.0010 Epoch 2/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 40s 248ms/step - loss: 0.7573 - recall: 0.5541 - val_loss: 0.8809 - val_recall: 0.5995 - learning_rate: 0.0010 Epoch 3/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 35s 243ms/step - loss: 0.6801 - recall: 0.5991 - val_loss: 0.8098 - val_recall: 0.6784 - learning_rate: 0.0010 Epoch 4/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 40s 233ms/step - loss: 0.5949 - recall: 0.6555 - val_loss: 0.9510 - val_recall: 0.6319 - learning_rate: 0.0010 Epoch 5/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 36s 239ms/step - loss: 0.5358 - recall: 0.6888 - val_loss: 0.8406 - val_recall: 0.6687 - learning_rate: 0.0010 Epoch 6/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 35s 243ms/step - loss: 0.5175 - recall: 0.7039 - val_loss: 0.7385 - val_recall: 0.6950 - learning_rate: 0.0010 Epoch 7/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 235ms/step - loss: 0.5096 - recall: 0.7264 - val_loss: 0.8432 - val_recall: 0.7108 - learning_rate: 0.0010 Epoch 8/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 48s 336ms/step - loss: 0.5263 - recall: 0.7275 - val_loss: 0.7060 - val_recall: 0.7432 - learning_rate: 0.0010 Epoch 9/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 33s 232ms/step - loss: 0.4338 - recall: 0.7747 - val_loss: 0.8316 - val_recall: 0.7546 - learning_rate: 0.0010 Epoch 10/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 239ms/step - loss: 0.4617 - recall: 0.7647 - val_loss: 0.8108 - val_recall: 0.7432 - learning_rate: 0.0010 Epoch 11/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 233ms/step - loss: 0.4197 - recall: 0.7834 - val_loss: 0.8501 - val_recall: 0.7406 - learning_rate: 0.0010 Epoch 12/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 237ms/step - loss: 0.4121 - recall: 0.7925 - val_loss: 0.7721 - val_recall: 0.7634 - learning_rate: 1.0000e-04 Epoch 13/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 33s 232ms/step - loss: 0.3817 - recall: 0.8064 - val_loss: 0.7482 - val_recall: 0.7713 - learning_rate: 1.0000e-04 Epoch 14/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 78s 545ms/step - loss: 0.3763 - recall: 0.8102 - val_loss: 0.7683 - val_recall: 0.7634 - learning_rate: 1.0000e-04 Epoch 15/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 236ms/step - loss: 0.3781 - recall: 0.7994 - val_loss: 0.7877 - val_recall: 0.7642 - learning_rate: 1.0000e-05 Epoch 16/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 34s 237ms/step - loss: 0.3701 - recall: 0.8088 - val_loss: 0.7936 - val_recall: 0.7642 - learning_rate: 1.0000e-05 Epoch 17/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 36s 248ms/step - loss: 0.3933 - recall: 0.8056 - val_loss: 0.7841 - val_recall: 0.7660 - learning_rate: 1.0000e-05 Epoch 18/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 35s 244ms/step - loss: 0.3852 - recall: 0.7920 - val_loss: 0.7832 - val_recall: 0.7660 - learning_rate: 1.0000e-06
In [152]:
##################################
# Evaluating the model
# for a simple CNN with dropout regularization
# on the independent validation set
##################################
model_dr_simple_y_pred_val = model_dr_simple.predict(val_gen)
36/36 ━━━━━━━━━━━━━━━━━━━━ 4s 112ms/step
In [153]:
##################################
# Plotting the loss profile
# for a simple CNN with dropout regularization
# on the training and validation sets
##################################
plot_training_history(model_dr_simple_history, 'Simple CNN With Dropout Regularization : ')
In [154]:
##################################
# Consolidating the predictions
# for a simple CNN with dropout regularization
# on the validation set
##################################
model_dr_simple_predictions_val = np.array(list(map(lambda x: np.argmax(x), model_dr_simple_y_pred_val)))
model_dr_simple_y_true_val = val_gen.classes
##################################
# Formulating the confusion matrix
# for a simple CNN with dropout regularization
# on the validation set
##################################
cmatrix_val = pd.DataFrame(confusion_matrix(model_dr_simple_y_true_val, model_dr_simple_predictions_val), columns=classes, index =classes)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with dropout regularization
# for each category of the validation set
##################################
plt.figure(figsize=(10, 6))
ax = sns.heatmap(cmatrix_val, annot = True, fmt = 'g' ,vmin = 0, vmax = 250, cmap = 'icefire')
ax.set_xlabel('Predicted',fontsize = 14,weight = 'bold')
ax.set_xticklabels(ax.get_xticklabels(),rotation =0)
ax.set_ylabel('Actual',fontsize = 14,weight = 'bold')
ax.set_yticklabels(ax.get_yticklabels(),rotation =0)
ax.set_title('Simple CNN With Dropout Regularization : Validation Set Confusion Matrix',fontsize = 14, weight = 'bold', pad=20);
##################################
# Resetting all states generated by Keras
##################################
keras.backend.clear_session()
In [155]:
##################################
# Calculating the model accuracy
# for a simple CNN with dropout regularization
# for the entire validation set
##################################
model_dr_simple_acc_val = accuracy_score(model_dr_simple_y_true_val, model_dr_simple_predictions_val)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with dropout regularization
# for the entire validation set
##################################
model_dr_simple_results_all_val = precision_recall_fscore_support(model_dr_simple_y_true_val, model_dr_simple_predictions_val, average='macro',zero_division = 1)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with dropout regularization
# for each category of the validation set
##################################
model_dr_simple_results_class_val = precision_recall_fscore_support(model_dr_simple_y_true_val, model_dr_simple_predictions_val, average=None, zero_division = 1)
##################################
# Consolidating all model evaluation metrics
# for a simple CNN with dropout regularization
##################################
metric_columns = ['Precision','Recall', 'F-Score','Support']
model_dr_simple_all_df_val = pd.concat([pd.DataFrame(list(model_dr_simple_results_class_val)).T,pd.DataFrame(list(model_dr_simple_results_all_val)).T])
model_dr_simple_all_df_val.columns = metric_columns
model_dr_simple_all_df_val.index = ['No Tumor', 'Glioma', 'Meningioma', 'Pituitary', 'Total']
print('Simple CNN With Dropout Regularization : Validation Set Classification Performance')
model_dr_simple_all_df_val
Simple CNN With Dropout Regularization : Validation Set Classification Performance
Out[155]:
| Precision | Recall | F-Score | Support | |
|---|---|---|---|---|
| No Tumor | 0.852459 | 0.815047 | 0.833333 | 319.0 |
| Glioma | 0.902778 | 0.738636 | 0.812500 | 264.0 |
| Meningioma | 0.554054 | 0.614232 | 0.582593 | 267.0 |
| Pituitary | 0.774691 | 0.862543 | 0.816260 | 291.0 |
| Total | 0.770996 | 0.757615 | 0.761172 | NaN |
In [156]:
##################################
# Consolidating all model evaluation metrics
# for a simple CNN with dropout regularization
##################################
model_dr_simple_model_list_val = []
model_dr_simple_measure_list_val = []
model_dr_simple_category_list_val = []
model_dr_simple_value_list_val = []
model_dr_simple_dataset_list_val = []
for i in range(3):
for j in range(5):
model_dr_simple_model_list_val.append('CNN_DR_Simple')
model_dr_simple_measure_list_val.append(metric_columns[i])
model_dr_simple_category_list_val.append(model_dr_simple_all_df_val.index[j])
model_dr_simple_value_list_val.append(model_dr_simple_all_df_val.iloc[j,i])
model_dr_simple_dataset_list_val.append('Validation')
model_dr_simple_all_summary_val = pd.DataFrame(zip(model_dr_simple_model_list_val,
model_dr_simple_measure_list_val,
model_dr_simple_category_list_val,
model_dr_simple_value_list_val,
model_dr_simple_dataset_list_val),
columns=['CNN.Model.Name',
'Model.Metric',
'Image.Category',
'Metric.Value',
'Data.Set'])
In [157]:
##################################
# Formulating the network architecture
# for a complex CNN with dropout regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_dr_complex = Sequential(name="model_dr_complex")
model_dr_complex.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="dr_complex_conv2d_0"))
model_dr_complex.add(MaxPooling2D(pool_size=(2, 2), name="dr_complex_max_pooling2d_0"))
model_dr_complex.add(Conv2D(filters=32, kernel_size=(3, 3), padding = 'Same', activation='relu', name="dr_complex_conv2d_1"))
model_dr_complex.add(MaxPooling2D(pool_size=(2, 2), name="dr_complex_max_pooling2d_1"))
model_dr_complex.add(Conv2D(filters=64, kernel_size=(3, 3), padding = 'Same', activation='relu', name="dr_complex_conv2d_2"))
model_dr_complex.add(MaxPooling2D(pool_size=(2, 2), name="dr_complex_max_pooling2d_2"))
model_dr_complex.add(Flatten(name="dr_complex_flatten"))
model_dr_complex.add(Dense(units=128, activation='relu', name="dr_complex_dense_0"))
model_dr_complex.add(Dropout(rate=0.30, name="dr_complex_dropout"))
model_dr_complex.add(Dense(units=num_classes, activation='softmax', name="dr_complex_dense_1"))
##################################
# Compiling the network layers
##################################
model_dr_complex.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [158]:
##################################
# Fitting the model
# for a complex CNN with dropout regularization
##################################
epochs = 20
set_seed()
model_dr_complex_history = model_dr_complex.fit(train_gen,
steps_per_epoch=len(train_gen)+1,
validation_steps=len(val_gen)+1,
validation_data=val_gen,
epochs=epochs,
verbose=1,
callbacks=[early_stopping, reduce_lr, dr_complex_model_checkpoint])
Epoch 1/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 371ms/step - loss: 1.0131 - recall: 0.3707 - val_loss: 0.8088 - val_recall: 0.6994 - learning_rate: 0.0010 Epoch 2/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 80s 360ms/step - loss: 0.4345 - recall: 0.8110 - val_loss: 0.7967 - val_recall: 0.6968 - learning_rate: 0.0010 Epoch 3/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 83s 364ms/step - loss: 0.2910 - recall: 0.8898 - val_loss: 0.7494 - val_recall: 0.7458 - learning_rate: 0.0010 Epoch 4/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 81s 358ms/step - loss: 0.2426 - recall: 0.9008 - val_loss: 0.7891 - val_recall: 0.7511 - learning_rate: 0.0010 Epoch 5/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 53s 369ms/step - loss: 0.1822 - recall: 0.9304 - val_loss: 0.6271 - val_recall: 0.7844 - learning_rate: 0.0010 Epoch 6/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 81s 365ms/step - loss: 0.1632 - recall: 0.9328 - val_loss: 0.7265 - val_recall: 0.7774 - learning_rate: 0.0010 Epoch 7/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 362ms/step - loss: 0.1317 - recall: 0.9478 - val_loss: 0.8423 - val_recall: 0.7862 - learning_rate: 0.0010 Epoch 8/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 51s 356ms/step - loss: 0.1286 - recall: 0.9583 - val_loss: 0.8516 - val_recall: 0.8107 - learning_rate: 0.0010 Epoch 9/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 360ms/step - loss: 0.0860 - recall: 0.9707 - val_loss: 0.7973 - val_recall: 0.8124 - learning_rate: 1.0000e-04 Epoch 10/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 103s 720ms/step - loss: 0.0758 - recall: 0.9745 - val_loss: 0.8234 - val_recall: 0.8081 - learning_rate: 1.0000e-04 Epoch 11/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 358ms/step - loss: 0.0523 - recall: 0.9825 - val_loss: 0.8551 - val_recall: 0.8098 - learning_rate: 1.0000e-04 Epoch 12/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 360ms/step - loss: 0.0571 - recall: 0.9813 - val_loss: 0.8562 - val_recall: 0.8054 - learning_rate: 1.0000e-05 Epoch 13/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 359ms/step - loss: 0.0540 - recall: 0.9823 - val_loss: 0.8620 - val_recall: 0.8089 - learning_rate: 1.0000e-05 Epoch 14/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 363ms/step - loss: 0.0564 - recall: 0.9793 - val_loss: 0.8652 - val_recall: 0.8098 - learning_rate: 1.0000e-05 Epoch 15/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 52s 359ms/step - loss: 0.0538 - recall: 0.9812 - val_loss: 0.8655 - val_recall: 0.8098 - learning_rate: 1.0000e-06
In [159]:
##################################
# Evaluating the model
# for a complex CNN with dropout regularization
# on the independent validation set
##################################
model_dr_complex_y_pred_val = model_dr_complex.predict(val_gen)
36/36 ━━━━━━━━━━━━━━━━━━━━ 4s 117ms/step
In [160]:
##################################
# Plotting the loss profile
# for a complex CNN with dropout regularization
# on the training and validation sets
##################################
plot_training_history(model_dr_complex_history, 'Complex CNN With Dropout Regularization : ')
In [161]:
##################################
# Consolidating the predictions
# for a complex CNN with dropout regularization
# on the validation set
##################################
model_dr_complex_predictions_val = np.array(list(map(lambda x: np.argmax(x), model_dr_complex_y_pred_val)))
model_dr_complex_y_true_val = val_gen.classes
##################################
# Formulating the confusion matrix
# for a complex CNN with dropout regularization
# on the validation set
##################################
cmatrix_val = pd.DataFrame(confusion_matrix(model_dr_complex_y_true_val, model_dr_complex_predictions_val), columns=classes, index =classes)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with dropout regularization
# for each category of the validation set
##################################
plt.figure(figsize=(10, 6))
ax = sns.heatmap(cmatrix_val, annot = True, fmt = 'g' ,vmin = 0, vmax = 250, cmap = 'icefire')
ax.set_xlabel('Predicted',fontsize = 14,weight = 'bold')
ax.set_xticklabels(ax.get_xticklabels(),rotation =0)
ax.set_ylabel('Actual',fontsize = 14,weight = 'bold')
ax.set_yticklabels(ax.get_yticklabels(),rotation =0)
ax.set_title('Complex CNN With Dropout Regularization : Validation Set Confusion Matrix',fontsize = 14, weight = 'bold', pad=20);
##################################
# Resetting all states generated by Keras
##################################
keras.backend.clear_session()
In [162]:
##################################
# Calculating the model accuracy
# for a complex CNN with dropout regularization
# for the entire validation set
##################################
model_dr_complex_acc_val = accuracy_score(model_dr_complex_y_true_val, model_dr_complex_predictions_val)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with dropout regularization
# for the entire validation set
##################################
model_dr_complex_results_all_val = precision_recall_fscore_support(model_dr_complex_y_true_val, model_dr_complex_predictions_val, average='macro',zero_division = 1)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with dropout regularization
# for each category of the validation set
##################################
model_dr_complex_results_class_val = precision_recall_fscore_support(model_dr_complex_y_true_val, model_dr_complex_predictions_val, average=None, zero_division = 1)
##################################
# Consolidating all model evaluation metrics
# for a complex CNN with dropout regularization
##################################
metric_columns = ['Precision','Recall', 'F-Score','Support']
model_dr_complex_all_df_val = pd.concat([pd.DataFrame(list(model_dr_complex_results_class_val)).T,pd.DataFrame(list(model_dr_complex_results_all_val)).T])
model_dr_complex_all_df_val.columns = metric_columns
model_dr_complex_all_df_val.index = ['No Tumor', 'Glioma', 'Meningioma', 'Pituitary', 'Total']
print('Complex CNN With Dropout Regularization : Validation Set Classification Performance')
model_dr_complex_all_df_val
Complex CNN With Dropout Regularization : Validation Set Classification Performance
Out[162]:
| Precision | Recall | F-Score | Support | |
|---|---|---|---|---|
| No Tumor | 0.835913 | 0.846395 | 0.841121 | 319.0 |
| Glioma | 0.858238 | 0.848485 | 0.853333 | 264.0 |
| Meningioma | 0.641975 | 0.584270 | 0.611765 | 267.0 |
| Pituitary | 0.815287 | 0.879725 | 0.846281 | 291.0 |
| Total | 0.787853 | 0.789719 | 0.788125 | NaN |
In [163]:
##################################
# Consolidating all model evaluation metrics
# for a complex CNN with dropout regularization
##################################
model_dr_complex_model_list_val = []
model_dr_complex_measure_list_val = []
model_dr_complex_category_list_val = []
model_dr_complex_value_list_val = []
model_dr_complex_dataset_list_val = []
for i in range(3):
for j in range(5):
model_dr_complex_model_list_val.append('CNN_DR_Complex')
model_dr_complex_measure_list_val.append(metric_columns[i])
model_dr_complex_category_list_val.append(model_dr_complex_all_df_val.index[j])
model_dr_complex_value_list_val.append(model_dr_complex_all_df_val.iloc[j,i])
model_dr_complex_dataset_list_val.append('Validation')
model_dr_complex_all_summary_val = pd.DataFrame(zip(model_dr_complex_model_list_val,
model_dr_complex_measure_list_val,
model_dr_complex_category_list_val,
model_dr_complex_value_list_val,
model_dr_complex_dataset_list_val),
columns=['CNN.Model.Name',
'Model.Metric',
'Image.Category',
'Metric.Value',
'Data.Set'])
1.6.5 CNN With Batch Normalization Regularization Model Fitting | Hyperparameter Tuning | Validation ¶
- The simple model contained 1,607,076 trainable parameters broken down per layer as follows:
- Conv2D: bnr_simple_conv2d_0
- output size = 227x227x8
- number of parameters = 80
- MaxPooling2D: bnr_simple_max_pooling2d_0
- output size = 113x113x8
- number of parameters = 0
- Conv2D: bnr_simple_conv2d_1
- output size = 113x113x16
- number of parameters = 1,168
- BatchNormalization: bnr_simple_batch_normalization
- output size = 113x113x16
- number of parameters = 64
- Activation: bnr_simple_activation
- output size = 113x113x16
- number of parameters = 0
- MaxPooling2D: bnr_simple_max_pooling2d_1
- output size = 56x56x16
- number of parameters = 0
- Flatten: bnr_simple_flatten
- output size = 50,176
- number of parameters = 0
- Dense: bnr_simple_dense_0
- output size = 32
- number of parameters = 1,605,664
- Dense: bnr_simple_dense_1
- output size = 4
- number of parameters = 132
- Conv2D: bnr_simple_conv2d_0
- The complex model contained 6,446,596 trainable parameters broken down per layer as follows:
- Conv2D: bnr_complex_conv2d_0
- output size = 227x227x16
- number of parameters = 160
- MaxPooling2D: bnr_complex_max_pooling2d_0
- output size = 113x113x16
- number of parameters = 0
- Conv2D: bnr_complex_conv2d_1
- output size = 113x113x32
- number of parameters = 4,640
- MaxPooling2D: bnr_complex_max_pooling2d_1
- output size = 56x56x32
- number of parameters = 0
- Conv2D: bnr_complex_conv2d_2
- output size = 56x56x64
- number of parameters = 18,496
- BatchNormalization: bnr_complex_batch_normalization
- output size = 56x56x64
- number of parameters = 256
- Activation: bnr_complex_activation
- output size = 56x56x64
- number of parameters = 0
- MaxPooling2D: bnr_complex_max_pooling2d_2
- output size = 28x28x64
- number of parameters = 0
- Flatten: bnr_complex_flatten
- output size = 50,176
- number of parameters = 0
- Dense: bnr_complex_dense_0
- output size = 128
- number of parameters = 6,422,656
- Dense: bnr_complex_dense_1
- output size = 4
- number of parameters = 516
- Conv2D: bnr_complex_conv2d_0
- The model performance on the validation set for all image categories is summarized as follows:
- Simple
- Precision = 0.8326
- Recall = 0.8261
- F1 Score = 0.8285
- Complex
- Precision = 0.6667
- Recall = 0.6539
- F1 Score = 0.6482
- Simple
In [164]:
##################################
# Formulating the network architecture
# for a simple CNN with batch normalization regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_bnr_simple = Sequential(name="model_bnr_simple")
model_bnr_simple.add(Conv2D(filters=8, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="bnr_simple_conv2d_0"))
model_bnr_simple.add(MaxPooling2D(pool_size=(2, 2), name="bnr_simple_max_pooling2d_0"))
model_bnr_simple.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', name="bnr_simple_conv2d_1"))
model_bnr_simple.add(BatchNormalization(name="bnr_simple_batch_normalization"))
model_bnr_simple.add(Activation('relu', name="bnr_simple_activation"))
model_bnr_simple.add(MaxPooling2D(pool_size=(2, 2), name="bnr_simple_max_pooling2d_1"))
model_bnr_simple.add(Flatten(name="bnr_simple_flatten"))
model_bnr_simple.add(Dense(units=32, activation='relu', name="bnr_simple_dense_0"))
model_bnr_simple.add(Dense(units=num_classes, activation='softmax', name="bnr_simple_dense_1"))
##################################
# Compiling the network layers
##################################
model_bnr_simple.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [165]:
##################################
# Fitting the model
# for a simple CNN with batch normalization regularization
##################################
epochs = 20
set_seed()
model_bnr_simple_history = model_bnr_simple.fit(train_gen,
steps_per_epoch=len(train_gen)+1,
validation_steps=len(val_gen)+1,
validation_data=val_gen,
epochs=epochs,
verbose=1,
callbacks=[early_stopping, reduce_lr, bnr_simple_model_checkpoint])
Epoch 1/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 42s 284ms/step - loss: 1.7668 - recall: 0.5558 - val_loss: 1.0888 - val_recall: 0.0473 - learning_rate: 0.0010 Epoch 2/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 39s 270ms/step - loss: 0.3585 - recall: 0.8676 - val_loss: 0.8608 - val_recall: 0.3716 - learning_rate: 0.0010 Epoch 3/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 44s 288ms/step - loss: 0.2334 - recall: 0.9148 - val_loss: 0.7054 - val_recall: 0.6591 - learning_rate: 0.0010 Epoch 4/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 79s 267ms/step - loss: 0.2119 - recall: 0.9237 - val_loss: 0.5743 - val_recall: 0.8089 - learning_rate: 0.0010 Epoch 5/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 41s 267ms/step - loss: 0.2065 - recall: 0.9280 - val_loss: 0.6802 - val_recall: 0.8072 - learning_rate: 0.0010 Epoch 6/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 39s 272ms/step - loss: 0.1448 - recall: 0.9461 - val_loss: 0.8415 - val_recall: 0.8387 - learning_rate: 0.0010 Epoch 7/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 40s 268ms/step - loss: 0.1309 - recall: 0.9561 - val_loss: 1.1974 - val_recall: 0.8107 - learning_rate: 0.0010 Epoch 8/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 39s 269ms/step - loss: 0.0820 - recall: 0.9706 - val_loss: 0.9800 - val_recall: 0.8282 - learning_rate: 1.0000e-04 Epoch 9/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 38s 266ms/step - loss: 0.0649 - recall: 0.9801 - val_loss: 1.0222 - val_recall: 0.8309 - learning_rate: 1.0000e-04 Epoch 10/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 39s 270ms/step - loss: 0.0683 - recall: 0.9782 - val_loss: 1.0025 - val_recall: 0.8247 - learning_rate: 1.0000e-04 Epoch 11/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 39s 268ms/step - loss: 0.0509 - recall: 0.9825 - val_loss: 0.9991 - val_recall: 0.8309 - learning_rate: 1.0000e-05 Epoch 12/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 68s 477ms/step - loss: 0.0648 - recall: 0.9745 - val_loss: 0.9882 - val_recall: 0.8309 - learning_rate: 1.0000e-05 Epoch 13/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 53s 275ms/step - loss: 0.0472 - recall: 0.9859 - val_loss: 0.9759 - val_recall: 0.8300 - learning_rate: 1.0000e-05 Epoch 14/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 39s 270ms/step - loss: 0.0499 - recall: 0.9851 - val_loss: 0.9774 - val_recall: 0.8309 - learning_rate: 1.0000e-06
In [166]:
##################################
# Evaluating the model
# for a simple CNN with batch normalization regularization
# on the independent validation set
##################################
model_bnr_simple_y_pred_val = model_bnr_simple.predict(val_gen)
36/36 ━━━━━━━━━━━━━━━━━━━━ 4s 110ms/step
In [167]:
##################################
# Plotting the loss profile
# for a simple CNN with batch normalization regularization
# on the training and validation sets
##################################
plot_training_history(model_bnr_simple_history, 'Simple CNN With Batch Normalization Regularization : ')
In [168]:
##################################
# Consolidating the predictions
# for a simple CNN with batch normalization regularization
# on the validation set
##################################
model_bnr_simple_predictions_val = np.array(list(map(lambda x: np.argmax(x), model_bnr_simple_y_pred_val)))
model_bnr_simple_y_true_val = val_gen.classes
##################################
# Formulating the confusion matrix
# for a simple CNN with batch normalization regularization
# on the validation set
##################################
cmatrix_val = pd.DataFrame(confusion_matrix(model_bnr_simple_y_true_val, model_bnr_simple_predictions_val), columns=classes, index =classes)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with batch normalization regularization
# for each category of the validation set
##################################
plt.figure(figsize=(10, 6))
ax = sns.heatmap(cmatrix_val, annot = True, fmt = 'g' ,vmin = 0, vmax = 250, cmap = 'icefire')
ax.set_xlabel('Predicted',fontsize = 14,weight = 'bold')
ax.set_xticklabels(ax.get_xticklabels(),rotation =0)
ax.set_ylabel('Actual',fontsize = 14,weight = 'bold')
ax.set_yticklabels(ax.get_yticklabels(),rotation =0)
ax.set_title('Simple CNN With Batch Normalization Regularization : Validation Set Confusion Matrix',fontsize = 14, weight = 'bold', pad=20);
##################################
# Resetting all states generated by Keras
##################################
keras.backend.clear_session()
In [169]:
##################################
# Calculating the model accuracy
# for a simple CNN with batch normalization regularization
# for the entire validation set
##################################
model_bnr_simple_acc_val = accuracy_score(model_bnr_simple_y_true_val, model_bnr_simple_predictions_val)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with batch normalization regularization
# for the entire validation set
##################################
model_bnr_simple_results_all_val = precision_recall_fscore_support(model_bnr_simple_y_true_val, model_bnr_simple_predictions_val, average='macro', zero_division = 1)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with batch normalization regularization
# for each category of the validation set
##################################
model_bnr_simple_results_class_val = precision_recall_fscore_support(model_bnr_simple_y_true_val, model_bnr_simple_predictions_val, average=None, zero_division = 1)
##################################
# Consolidating all model evaluation metrics
# for a simple CNN with batch normalization regularization
##################################
metric_columns = ['Precision','Recall', 'F-Score','Support']
model_bnr_simple_all_df_val = pd.concat([pd.DataFrame(list(model_bnr_simple_results_class_val)).T,pd.DataFrame(list(model_bnr_simple_results_all_val)).T])
model_bnr_simple_all_df_val.columns = metric_columns
model_bnr_simple_all_df_val.index = ['No Tumor', 'Glioma', 'Meningioma', 'Pituitary', 'Total']
print('Simple CNN With Batch Normalization Regularization : Validation Set Classification Performance')
model_bnr_simple_all_df_val
Simple CNN With Batch Normalization Regularization : Validation Set Classification Performance
Out[169]:
| Precision | Recall | F-Score | Support | |
|---|---|---|---|---|
| No Tumor | 0.832317 | 0.855799 | 0.843895 | 319.0 |
| Glioma | 0.962025 | 0.863636 | 0.910180 | 264.0 |
| Meningioma | 0.690647 | 0.719101 | 0.704587 | 267.0 |
| Pituitary | 0.845638 | 0.865979 | 0.855688 | 291.0 |
| Total | 0.832657 | 0.826129 | 0.828587 | NaN |
In [170]:
##################################
# Consolidating all model evaluation metrics
# for a simple CNN with batch normalization regularization
##################################
model_bnr_simple_model_list_val = []
model_bnr_simple_measure_list_val = []
model_bnr_simple_category_list_val = []
model_bnr_simple_value_list_val = []
model_bnr_simple_dataset_list_val = []
for i in range(3):
for j in range(5):
model_bnr_simple_model_list_val.append('CNN_BNR_Simple')
model_bnr_simple_measure_list_val.append(metric_columns[i])
model_bnr_simple_category_list_val.append(model_bnr_simple_all_df_val.index[j])
model_bnr_simple_value_list_val.append(model_bnr_simple_all_df_val.iloc[j,i])
model_bnr_simple_dataset_list_val.append('Validation')
model_bnr_simple_all_summary_val = pd.DataFrame(zip(model_bnr_simple_model_list_val,
model_bnr_simple_measure_list_val,
model_bnr_simple_category_list_val,
model_bnr_simple_value_list_val,
model_bnr_simple_dataset_list_val),
columns=['CNN.Model.Name',
'Model.Metric',
'Image.Category',
'Metric.Value',
'Data.Set'])
In [171]:
##################################
# Formulating the network architecture
# for a complex CNN with batch normalization regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_bnr_complex = Sequential(name="model_bnr_complex")
model_bnr_complex.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="bnr_complex_conv2d_0"))
model_bnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="bnr_complex_max_pooling2d_0"))
model_bnr_complex.add(Conv2D(filters=32, kernel_size=(3, 3), padding = 'Same', activation='relu', name="bnr_complex_conv2d_1"))
model_bnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="bnr_complex_max_pooling2d_1"))
model_bnr_complex.add(Conv2D(filters=64, kernel_size=(3, 3), padding = 'Same', activation='relu', name="bnr_complex_conv2d_2"))
model_bnr_complex.add(BatchNormalization(name="bnr_complex_batch_normalization"))
model_bnr_complex.add(Activation('relu', name="bnr_complex_activation"))
model_bnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="bnr_complex_max_pooling2d_2"))
model_bnr_complex.add(Flatten(name="bnr_complex_flatten"))
model_bnr_complex.add(Dense(units=128, activation='relu', name="bnr_complex_dense_0"))
model_bnr_complex.add(Dense(units=num_classes, activation='softmax', name="bnr_complex_dense_1"))
##################################
# Compiling the network layers
##################################
model_bnr_complex.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [172]:
##################################
# Fitting the model
# for a complex CNN with batch normalization regularization
##################################
epochs = 20
set_seed()
model_bnr_complex_history = model_bnr_complex.fit(train_gen,
steps_per_epoch=len(train_gen)+1,
validation_steps=len(val_gen)+1,
validation_data=val_gen,
epochs=epochs,
verbose=1,
callbacks=[early_stopping, reduce_lr, bnr_complex_model_checkpoint])
Epoch 1/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 61s 411ms/step - loss: 2.4198 - recall: 0.4782 - val_loss: 1.1481 - val_recall: 0.0096 - learning_rate: 0.0010 Epoch 2/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 79s 387ms/step - loss: 0.3966 - recall: 0.8304 - val_loss: 0.9454 - val_recall: 0.1613 - learning_rate: 0.0010 Epoch 3/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 81s 384ms/step - loss: 0.2384 - recall: 0.9055 - val_loss: 0.7357 - val_recall: 0.5819 - learning_rate: 0.0010 Epoch 4/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 83s 389ms/step - loss: 0.2179 - recall: 0.9136 - val_loss: 0.6788 - val_recall: 0.7809 - learning_rate: 0.0010 Epoch 5/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 81s 383ms/step - loss: 0.1693 - recall: 0.9332 - val_loss: 0.8541 - val_recall: 0.7064 - learning_rate: 0.0010 Epoch 6/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 385ms/step - loss: 0.1205 - recall: 0.9529 - val_loss: 0.8922 - val_recall: 0.7774 - learning_rate: 0.0010 Epoch 7/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 56s 387ms/step - loss: 0.1140 - recall: 0.9631 - val_loss: 1.1084 - val_recall: 0.7695 - learning_rate: 0.0010 Epoch 8/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 382ms/step - loss: 0.0670 - recall: 0.9783 - val_loss: 0.8778 - val_recall: 0.8151 - learning_rate: 1.0000e-04 Epoch 9/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 54s 377ms/step - loss: 0.0429 - recall: 0.9854 - val_loss: 0.8952 - val_recall: 0.8186 - learning_rate: 1.0000e-04 Epoch 10/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 381ms/step - loss: 0.0439 - recall: 0.9852 - val_loss: 0.8729 - val_recall: 0.8335 - learning_rate: 1.0000e-04
In [173]:
##################################
# Evaluating the model
# for a complex CNN with batch normalization regularization
# on the independent validation set
##################################
model_bnr_complex_y_pred_val = model_bnr_complex.predict(val_gen)
36/36 ━━━━━━━━━━━━━━━━━━━━ 5s 121ms/step
In [174]:
##################################
# Plotting the loss profile
# for a complex CNN with batch normalization regularization
# on the training and validation sets
##################################
plot_training_history(model_bnr_complex_history, 'Complex CNN With Batch Normalization Regularization : ')
In [175]:
##################################
# Consolidating the predictions
# for a complex CNN with batch normalization regularization
# on the validation set
##################################
model_bnr_complex_predictions_val = np.array(list(map(lambda x: np.argmax(x), model_bnr_complex_y_pred_val)))
model_bnr_complex_y_true_val = val_gen.classes
##################################
# Formulating the confusion matrix
# for a complex CNN with batch normalization regularization
# on the validation set
##################################
cmatrix_val = pd.DataFrame(confusion_matrix(model_bnr_complex_y_true_val, model_bnr_complex_predictions_val), columns=classes, index =classes)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with batch normalization regularization
# for each category of the validation set
##################################
plt.figure(figsize=(10, 6))
ax = sns.heatmap(cmatrix_val, annot = True, fmt = 'g' ,vmin = 0, vmax = 250, cmap = 'icefire')
ax.set_xlabel('Predicted',fontsize = 14,weight = 'bold')
ax.set_xticklabels(ax.get_xticklabels(),rotation =0)
ax.set_ylabel('Actual',fontsize = 14,weight = 'bold')
ax.set_yticklabels(ax.get_yticklabels(),rotation =0)
ax.set_title('Complex CNN With Batch Normalization Regularization : Validation Set Confusion Matrix',fontsize = 14, weight = 'bold', pad=20);
##################################
# Resetting all states generated by Keras
##################################
keras.backend.clear_session()
In [176]:
##################################
# Calculating the model accuracy
# for a complex CNN with batch normalization regularization
# for the entire validation set
##################################
model_bnr_complex_acc_val = accuracy_score(model_bnr_complex_y_true_val, model_bnr_complex_predictions_val)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with batch normalization regularization
# for the entire validation set
##################################
model_bnr_complex_results_all_val = precision_recall_fscore_support(model_bnr_complex_y_true_val, model_bnr_complex_predictions_val, average='macro', zero_division = 1)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with batch normalization regularization
# for each category of the validation set
##################################
model_bnr_complex_results_class_val = precision_recall_fscore_support(model_bnr_complex_y_true_val, model_bnr_complex_predictions_val, average=None, zero_division = 1)
##################################
# Consolidating all model evaluation metrics
# for a complex CNN with batch normalization regularization
##################################
metric_columns = ['Precision','Recall', 'F-Score','Support']
model_bnr_complex_all_df_val = pd.concat([pd.DataFrame(list(model_bnr_complex_results_class_val)).T,pd.DataFrame(list(model_bnr_complex_results_all_val)).T])
model_bnr_complex_all_df_val.columns = metric_columns
model_bnr_complex_all_df_val.index = ['No Tumor', 'Glioma', 'Meningioma', 'Pituitary', 'Total']
print('Complex CNN With Batch Normalization Regularization : Validation Set Classification Performance')
model_bnr_complex_all_df_val
Complex CNN With Batch Normalization Regularization : Validation Set Classification Performance
Out[176]:
| Precision | Recall | F-Score | Support | |
|---|---|---|---|---|
| No Tumor | 0.833333 | 0.611285 | 0.705244 | 319.0 |
| Glioma | 0.623684 | 0.897727 | 0.736025 | 264.0 |
| Meningioma | 0.430070 | 0.460674 | 0.444846 | 267.0 |
| Pituitary | 0.780083 | 0.646048 | 0.706767 | 291.0 |
| Total | 0.666793 | 0.653934 | 0.648221 | NaN |
In [177]:
##################################
# Consolidating all model evaluation metrics
# for a complex CNN with batch normalization regularization
##################################
model_bnr_complex_model_list_val = []
model_bnr_complex_measure_list_val = []
model_bnr_complex_category_list_val = []
model_bnr_complex_value_list_val = []
model_bnr_complex_dataset_list_val = []
for i in range(3):
for j in range(5):
model_bnr_complex_model_list_val.append('CNN_BNR_Complex')
model_bnr_complex_measure_list_val.append(metric_columns[i])
model_bnr_complex_category_list_val.append(model_bnr_complex_all_df_val.index[j])
model_bnr_complex_value_list_val.append(model_bnr_complex_all_df_val.iloc[j,i])
model_bnr_complex_dataset_list_val.append('Validation')
model_bnr_complex_all_summary_val = pd.DataFrame(zip(model_bnr_complex_model_list_val,
model_bnr_complex_measure_list_val,
model_bnr_complex_category_list_val,
model_bnr_complex_value_list_val,
model_bnr_complex_dataset_list_val),
columns=['CNN.Model.Name',
'Model.Metric',
'Image.Category',
'Metric.Value',
'Data.Set'])
1.6.6 CNN With Dropout and Batch Normalization Regularization Model Fitting | Hyperparameter Tuning | Validation ¶
- The simple model contained 1,607,076 trainable parameters broken down per layer as follows:
- Conv2D: cdrbnr_simple_conv2d_0
- output size = 227x227x8
- number of parameters = 80
- MaxPooling2D: cdrbnr_simple_max_pooling2d_0
- output size = 113x113x8
- number of parameters = 0
- Conv2D: cdrbnr_simple_conv2d_1
- output size = 113x113x16
- number of parameters = 1,168
- BatchNormalization: cdrbnr_simple_batch_normalization
- output size = 113x113x16
- number of parameters = 64
- Activation: cdrbnr_simple_activation
- output size = 113x113x16
- number of parameters = 0
- MaxPooling2D: cdrbnr_simple_max_pooling2d_1
- output size = 56x56x16
- number of parameters = 0
- Flatten: cdrbnr_simple_flatten
- output size = 50,176
- number of parameters = 0
- Dense: cdrbnr_simple_dense_0
- output size = 32
- number of parameters = 1,605,664
- Dropout: cdrbnr_simple_dropout
- output size = 32
- number of parameters = 0
- Dense: cdrbnr_simple_dense_1
- output size = 4
- number of parameters = 132
- Conv2D: cdrbnr_simple_conv2d_0
- The complex model contained 6,446,596 trainable parameters broken down per layer as follows:
- Conv2D: cdrbnr_complex_conv2d_0
- output size = 227x227x16
- number of parameters = 160
- MaxPooling2D: cdrbnr_complex_max_pooling2d_0
- output size = 113x113x16
- number of parameters = 0
- Conv2D: cdrbnr_complex_conv2d_1
- output size = 113x113x32
- number of parameters = 4,640
- MaxPooling2D: cdrbnr_complex_max_pooling2d_1
- output size = 56x56x32
- number of parameters = 0
- Conv2D: cdrbnr_complex_conv2d_2
- output size = 56x56x64
- number of parameters = 18,496
- BatchNormalization: cdrbnr_complex_batch_normalization
- output size = 56x56x64
- number of parameters = 256
- Activation: cdrbnr_complex_activation
- output size = 56x56x64
- number of parameters = 0
- MaxPooling2D: cdrbnr_complex_max_pooling2d_2
- output size = 28x28x64
- number of parameters = 0
- Flatten: cdrbnr_complex_flatten
- output size = 50,176
- number of parameters = 0
- Dense: cdrbnr_complex_dense_0
- output size = 128
- number of parameters = 6,422,656
- Dropout: cdrbnr_complex_dropout
- output size = 128
- number of parameters = 0
- Dense: cdrbnr_complex_dense_1
- output size = 4
- number of parameters = 516
- Conv2D: cdrbnr_complex_conv2d_0
- The model performance on the validation set for all image categories is summarized as follows:
- Simple
- Precision = 0.6003
- Recall = 0.4639
- F1 Score = 0.4181
- Complex
- Precision = 0.8429
- Recall = 0.8385
- F1 Score = 0.8388
- Simple
In [178]:
##################################
# Formulating the network architecture
# for a simple CNN with dropout and batch normalization regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_cdrbnr_simple = Sequential(name="model_cdrbnr_simple")
model_cdrbnr_simple.add(Conv2D(filters=8, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="cdrbnr_simple_conv2d_0"))
model_cdrbnr_simple.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_simple_max_pooling2d_0"))
model_cdrbnr_simple.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', name="cdrbnr_simple_conv2d_1"))
model_cdrbnr_simple.add(BatchNormalization(name="cdrbnr_simple_batch_normalization"))
model_cdrbnr_simple.add(Activation('relu', name="cdrbnr_simple_activation"))
model_cdrbnr_simple.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_simple_max_pooling2d_1"))
model_cdrbnr_simple.add(Flatten(name="cdrbnr_simple_flatten"))
model_cdrbnr_simple.add(Dense(units=32, activation='relu', name="cdrbnr_simple_dense_0"))
model_cdrbnr_simple.add(Dropout(rate=0.30, name="cdrbnr_simple_dropout"))
model_cdrbnr_simple.add(Dense(units=num_classes, activation='softmax', name="cdrbnr_simple_dense_1"))
##################################
# Compiling the network layers
##################################
model_cdrbnr_simple.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [179]:
##################################
# Fitting the model
# for a simple CNN with dropout and batch normalization regularization
##################################
epochs = 20
set_seed()
model_cdrbnr_simple_history = model_cdrbnr_simple.fit(train_gen,
steps_per_epoch=len(train_gen)+1,
validation_steps=len(val_gen)+1,
validation_data=val_gen,
epochs=epochs,
verbose=1,
callbacks=[early_stopping, reduce_lr, cdrbnr_simple_model_checkpoint])
Epoch 1/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 41s 274ms/step - loss: 1.6579 - recall: 0.1515 - val_loss: 1.3345 - val_recall: 0.0018 - learning_rate: 0.0010 Epoch 2/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 39s 271ms/step - loss: 1.0206 - recall: 0.3417 - val_loss: 1.1807 - val_recall: 0.0649 - learning_rate: 0.0010 Epoch 3/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 40s 264ms/step - loss: 0.9324 - recall: 0.3955 - val_loss: 1.0523 - val_recall: 0.2366 - learning_rate: 0.0010 Epoch 4/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 43s 277ms/step - loss: 0.7758 - recall: 0.4966 - val_loss: 0.9607 - val_recall: 0.4137 - learning_rate: 0.0010 Epoch 5/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 39s 265ms/step - loss: 0.7319 - recall: 0.5117 - val_loss: 1.0513 - val_recall: 0.4496 - learning_rate: 0.0010 Epoch 6/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 65s 455ms/step - loss: 0.6944 - recall: 0.5397 - val_loss: 1.0002 - val_recall: 0.5127 - learning_rate: 0.0010 Epoch 7/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 39s 268ms/step - loss: 0.6810 - recall: 0.5275 - val_loss: 1.1606 - val_recall: 0.6056 - learning_rate: 0.0010 Epoch 8/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 40s 275ms/step - loss: 0.6298 - recall: 0.5520 - val_loss: 0.9720 - val_recall: 0.5951 - learning_rate: 1.0000e-04 Epoch 9/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 38s 265ms/step - loss: 0.5942 - recall: 0.5613 - val_loss: 0.9829 - val_recall: 0.5960 - learning_rate: 1.0000e-04 Epoch 10/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 38s 262ms/step - loss: 0.6268 - recall: 0.5480 - val_loss: 1.0679 - val_recall: 0.5942 - learning_rate: 1.0000e-04
In [180]:
##################################
# Evaluating the model
# for a simple CNN with dropout and batch normalization regularization
# on the independent validation set
##################################
model_cdrbnr_simple_y_pred_val = model_cdrbnr_simple.predict(val_gen)
36/36 ━━━━━━━━━━━━━━━━━━━━ 4s 100ms/step
In [181]:
##################################
# Plotting the loss profile
# for a simple CNN with dropout and batch normalization regularization
# on the training and validation sets
##################################
plot_training_history(model_cdrbnr_simple_history, 'Simple CNN With Dropout and Batch Normalization Regularization : ')
In [182]:
##################################
# Consolidating the predictions
# for a simple CNN with dropout and batch normalization regularization
# on the validation set
##################################
model_cdrbnr_simple_predictions_val = np.array(list(map(lambda x: np.argmax(x), model_cdrbnr_simple_y_pred_val)))
model_cdrbnr_simple_y_true_val = val_gen.classes
##################################
# Formulating the confusion matrix
# for a simple CNN with dropout and batch normalization regularization
# on the validation set
##################################
cmatrix_val = pd.DataFrame(confusion_matrix(model_cdrbnr_simple_y_true_val, model_cdrbnr_simple_predictions_val), columns=classes, index =classes)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with dropout and batch normalization regularization
# for each category of the validation set
##################################
plt.figure(figsize=(10, 6))
ax = sns.heatmap(cmatrix_val, annot = True, fmt = 'g' ,vmin = 0, vmax = 250, cmap = 'icefire')
ax.set_xlabel('Predicted',fontsize = 14,weight = 'bold')
ax.set_xticklabels(ax.get_xticklabels(),rotation =0)
ax.set_ylabel('Actual',fontsize = 14,weight = 'bold')
ax.set_yticklabels(ax.get_yticklabels(),rotation =0)
ax.set_title('Simple CNN With Dropout and Batch Normalization Regularization : Validation Set Confusion Matrix',fontsize = 14, weight = 'bold', pad=20);
##################################
# Resetting all states generated by Keras
##################################
keras.backend.clear_session()
In [183]:
##################################
# Calculating the model accuracy
# for a simple CNN with dropout and batch normalization regularization
# for the entire validation set
##################################
model_cdrbnr_simple_acc_val = accuracy_score(model_cdrbnr_simple_y_true_val, model_cdrbnr_simple_predictions_val)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with dropout and batch normalization regularization
# for the entire validation set
##################################
model_cdrbnr_simple_results_all_val = precision_recall_fscore_support(model_cdrbnr_simple_y_true_val, model_cdrbnr_simple_predictions_val, average='macro', zero_division = 1)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a simple CNN with dropout and batch normalization regularization
# for each category of the validation set
##################################
model_cdrbnr_simple_results_class_val = precision_recall_fscore_support(model_cdrbnr_simple_y_true_val, model_cdrbnr_simple_predictions_val, average=None, zero_division = 1)
##################################
# Consolidating all model evaluation metrics
# for a simple CNN with dropout and batch normalization regularization
##################################
metric_columns = ['Precision','Recall', 'F-Score','Support']
model_cdrbnr_simple_all_df_val = pd.concat([pd.DataFrame(list(model_cdrbnr_simple_results_class_val)).T,pd.DataFrame(list(model_cdrbnr_simple_results_all_val)).T])
model_cdrbnr_simple_all_df_val.columns = metric_columns
model_cdrbnr_simple_all_df_val.index = ['No Tumor', 'Glioma', 'Meningioma', 'Pituitary', 'Total']
print('Simple CNN With Dropout and Batch Normalization Regularization : Validation Set Classification Performance')
model_cdrbnr_simple_all_df_val
Simple CNN With Dropout and Batch Normalization Regularization : Validation Set Classification Performance
Out[183]:
| Precision | Recall | F-Score | Support | |
|---|---|---|---|---|
| No Tumor | 0.722656 | 0.579937 | 0.643478 | 319.0 |
| Glioma | 0.875000 | 0.026515 | 0.051471 | 264.0 |
| Meningioma | 0.313199 | 0.524345 | 0.392157 | 267.0 |
| Pituitary | 0.490698 | 0.725086 | 0.585298 | 291.0 |
| Total | 0.600388 | 0.463971 | 0.418101 | NaN |
In [184]:
##################################
# Consolidating all model evaluation metrics
# for a simple CNN with dropout and batch normalization regularization
##################################
model_cdrbnr_simple_model_list_val = []
model_cdrbnr_simple_measure_list_val = []
model_cdrbnr_simple_category_list_val = []
model_cdrbnr_simple_value_list_val = []
model_cdrbnr_simple_dataset_list_val = []
for i in range(3):
for j in range(5):
model_cdrbnr_simple_model_list_val.append('CNN_CDRBNR_Simple')
model_cdrbnr_simple_measure_list_val.append(metric_columns[i])
model_cdrbnr_simple_category_list_val.append(model_cdrbnr_simple_all_df_val.index[j])
model_cdrbnr_simple_value_list_val.append(model_cdrbnr_simple_all_df_val.iloc[j,i])
model_cdrbnr_simple_dataset_list_val.append('Validation')
model_cdrbnr_simple_all_summary_val = pd.DataFrame(zip(model_cdrbnr_simple_model_list_val,
model_cdrbnr_simple_measure_list_val,
model_cdrbnr_simple_category_list_val,
model_cdrbnr_simple_value_list_val,
model_cdrbnr_simple_dataset_list_val),
columns=['CNN.Model.Name',
'Model.Metric',
'Image.Category',
'Metric.Value',
'Data.Set'])
In [185]:
##################################
# Formulating the network architecture
# for a complex CNN with dropout and batch normalization regularization
##################################
set_seed()
batch_size = 32
input_shape = (227, 227, 1)
model_cdrbnr_complex = Sequential(name="model_cdrbnr_complex")
model_cdrbnr_complex.add(Conv2D(filters=16, kernel_size=(3, 3), padding = 'Same', activation='relu', input_shape=(227, 227, 1), name="cdrbnr_complex_conv2d_0"))
model_cdrbnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_complex_max_pooling2d_0"))
model_cdrbnr_complex.add(Conv2D(filters=32, kernel_size=(3, 3), padding = 'Same', activation='relu', name="cdrbnr_complex_conv2d_1"))
model_cdrbnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_complex_max_pooling2d_1"))
model_cdrbnr_complex.add(Conv2D(filters=64, kernel_size=(3, 3), padding = 'Same', activation='relu', name="cdrbnr_complex_conv2d_2"))
model_cdrbnr_complex.add(BatchNormalization(name="cdrbnr_complex_batch_normalization"))
model_cdrbnr_complex.add(Activation('relu', name="cdrbnr_complex_activation"))
model_cdrbnr_complex.add(MaxPooling2D(pool_size=(2, 2), name="cdrbnr_complex_max_pooling2d_2"))
model_cdrbnr_complex.add(Flatten(name="cdrbnr_complex_flatten"))
model_cdrbnr_complex.add(Dense(units=128, activation='relu', name="cdrbnr_complex_dense_0"))
model_cdrbnr_complex.add(Dropout(rate=0.30, name="cdrbnr_complex_dropout"))
model_cdrbnr_complex.add(Dense(units=num_classes, activation='softmax', name="cdrbnr_complex_dense_1"))
##################################
# Compiling the network layers
##################################
model_cdrbnr_complex.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [186]:
##################################
# Fitting the model
# for a complex CNN with dropout and batch normalization regularization
##################################
epochs = 20
set_seed()
model_cdrbnr_complex_history = model_cdrbnr_complex.fit(train_gen,
steps_per_epoch=len(train_gen)+1,
validation_steps=len(val_gen)+1,
validation_data=val_gen,
epochs=epochs,
verbose=1,
callbacks=[early_stopping, reduce_lr, cdrbnr_complex_model_checkpoint])
Epoch 1/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 57s 382ms/step - loss: 1.7995 - recall: 0.5219 - val_loss: 1.1321 - val_recall: 0.0342 - learning_rate: 0.0010 Epoch 2/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 83s 389ms/step - loss: 0.3938 - recall: 0.8333 - val_loss: 0.9887 - val_recall: 0.0649 - learning_rate: 0.0010 Epoch 3/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 82s 388ms/step - loss: 0.2484 - recall: 0.8988 - val_loss: 0.6290 - val_recall: 0.6713 - learning_rate: 0.0010 Epoch 4/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 82s 388ms/step - loss: 0.2268 - recall: 0.9093 - val_loss: 0.6252 - val_recall: 0.7555 - learning_rate: 0.0010 Epoch 5/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 81s 382ms/step - loss: 0.1590 - recall: 0.9359 - val_loss: 0.8430 - val_recall: 0.7046 - learning_rate: 0.0010 Epoch 6/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 56s 387ms/step - loss: 0.1436 - recall: 0.9409 - val_loss: 0.5680 - val_recall: 0.8352 - learning_rate: 0.0010 Epoch 7/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 56s 390ms/step - loss: 0.1110 - recall: 0.9563 - val_loss: 0.7335 - val_recall: 0.8344 - learning_rate: 0.0010 Epoch 8/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 380ms/step - loss: 0.1024 - recall: 0.9607 - val_loss: 0.9613 - val_recall: 0.8291 - learning_rate: 0.0010 Epoch 9/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 382ms/step - loss: 0.1047 - recall: 0.9615 - val_loss: 0.6784 - val_recall: 0.8475 - learning_rate: 0.0010 Epoch 10/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 381ms/step - loss: 0.0612 - recall: 0.9779 - val_loss: 0.7055 - val_recall: 0.8580 - learning_rate: 1.0000e-04 Epoch 11/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 56s 387ms/step - loss: 0.0465 - recall: 0.9825 - val_loss: 0.7504 - val_recall: 0.8615 - learning_rate: 1.0000e-04 Epoch 12/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 57s 393ms/step - loss: 0.0426 - recall: 0.9825 - val_loss: 0.8035 - val_recall: 0.8624 - learning_rate: 1.0000e-04 Epoch 13/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 56s 387ms/step - loss: 0.0373 - recall: 0.9885 - val_loss: 0.7971 - val_recall: 0.8624 - learning_rate: 1.0000e-05 Epoch 14/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 385ms/step - loss: 0.0427 - recall: 0.9842 - val_loss: 0.7896 - val_recall: 0.8606 - learning_rate: 1.0000e-05 Epoch 15/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 55s 381ms/step - loss: 0.0323 - recall: 0.9903 - val_loss: 0.7911 - val_recall: 0.8606 - learning_rate: 1.0000e-05 Epoch 16/20 144/144 ━━━━━━━━━━━━━━━━━━━━ 56s 391ms/step - loss: 0.0418 - recall: 0.9818 - val_loss: 0.7901 - val_recall: 0.8606 - learning_rate: 1.0000e-06
In [187]:
##################################
# Evaluating the model
# for a complex CNN with dropout and batch normalization regularization
# on the independent validation set
##################################
model_cdrbnr_complex_y_pred_val = model_cdrbnr_complex.predict(val_gen)
36/36 ━━━━━━━━━━━━━━━━━━━━ 5s 123ms/step
In [188]:
##################################
# Plotting the loss profile
# for a complex CNN with dropout and batch normalization regularization
# on the training and validation sets
##################################
plot_training_history(model_cdrbnr_complex_history, 'Complex CNN With Dropout and Batch Normalization Regularization : ')
In [189]:
##################################
# Consolidating the predictions
# for a complex CNN with dropout and batch normalization regularization
# on the validation set
##################################
model_cdrbnr_complex_predictions_val = np.array(list(map(lambda x: np.argmax(x), model_cdrbnr_complex_y_pred_val)))
model_cdrbnr_complex_y_true_val = val_gen.classes
##################################
# Formulating the confusion matrix
# for a complex CNN with dropout and batch normalization regularization
# on the validation set
##################################
cmatrix_val = pd.DataFrame(confusion_matrix(model_cdrbnr_complex_y_true_val, model_cdrbnr_complex_predictions_val), columns=classes, index =classes)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with dropout and batch normalization regularization
# for each category of the validation set
##################################
plt.figure(figsize=(10, 6))
ax = sns.heatmap(cmatrix_val, annot = True, fmt = 'g' ,vmin = 0, vmax = 250, cmap = 'icefire')
ax.set_xlabel('Predicted',fontsize = 14,weight = 'bold')
ax.set_xticklabels(ax.get_xticklabels(),rotation =0)
ax.set_ylabel('Actual',fontsize = 14,weight = 'bold')
ax.set_yticklabels(ax.get_yticklabels(),rotation =0)
ax.set_title('Complex CNN With Dropout and Batch Normalization Regularization : Validation Set Confusion Matrix',fontsize = 14, weight = 'bold', pad=20);
##################################
# Resetting all states generated by Keras
##################################
keras.backend.clear_session()
In [190]:
##################################
# Calculating the model accuracy
# for a complex CNN with dropout and batch normalization regularization
# for the entire validation set
##################################
model_cdrbnr_complex_acc_val = accuracy_score(model_cdrbnr_complex_y_true_val, model_cdrbnr_complex_predictions_val)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with dropout and batch normalization regularization
# for the entire validation set
##################################
model_cdrbnr_complex_results_all_val = precision_recall_fscore_support(model_cdrbnr_complex_y_true_val, model_cdrbnr_complex_predictions_val, average='macro', zero_division = 1)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with dropout and batch normalization regularization
# for each category of the validation set
##################################
model_cdrbnr_complex_results_class_val = precision_recall_fscore_support(model_cdrbnr_complex_y_true_val, model_cdrbnr_complex_predictions_val, average=None, zero_division = 1)
##################################
# Consolidating all model evaluation metrics
# for a complex CNN with dropout and batch normalization regularization
##################################
metric_columns = ['Precision','Recall', 'F-Score','Support']
model_cdrbnr_complex_all_df_val = pd.concat([pd.DataFrame(list(model_cdrbnr_complex_results_class_val)).T,pd.DataFrame(list(model_cdrbnr_complex_results_all_val)).T])
model_cdrbnr_complex_all_df_val.columns = metric_columns
model_cdrbnr_complex_all_df_val.index = ['No Tumor', 'Glioma', 'Meningioma', 'Pituitary', 'Total']
print('Complex CNN With Dropout and Batch Normalization Regularization : Validation Set Classification Performance')
model_cdrbnr_complex_all_df_val
Complex CNN With Dropout and Batch Normalization Regularization : Validation Set Classification Performance
Out[190]:
| Precision | Recall | F-Score | Support | |
|---|---|---|---|---|
| No Tumor | 0.848765 | 0.862069 | 0.855365 | 319.0 |
| Glioma | 0.923077 | 0.818182 | 0.867470 | 264.0 |
| Meningioma | 0.753968 | 0.711610 | 0.732177 | 267.0 |
| Pituitary | 0.845921 | 0.962199 | 0.900322 | 291.0 |
| Total | 0.842933 | 0.838515 | 0.838834 | NaN |
In [191]:
##################################
# Consolidating all model evaluation metrics
# for a complex CNN with dropout and batch normalization regularization
##################################
model_cdrbnr_complex_model_list_val = []
model_cdrbnr_complex_measure_list_val = []
model_cdrbnr_complex_category_list_val = []
model_cdrbnr_complex_value_list_val = []
model_cdrbnr_complex_dataset_list_val = []
for i in range(3):
for j in range(5):
model_cdrbnr_complex_model_list_val.append('CNN_CDRBNR_Complex')
model_cdrbnr_complex_measure_list_val.append(metric_columns[i])
model_cdrbnr_complex_category_list_val.append(model_cdrbnr_complex_all_df_val.index[j])
model_cdrbnr_complex_value_list_val.append(model_cdrbnr_complex_all_df_val.iloc[j,i])
model_cdrbnr_complex_dataset_list_val.append('Validation')
model_cdrbnr_complex_all_summary_val = pd.DataFrame(zip(model_cdrbnr_complex_model_list_val,
model_cdrbnr_complex_measure_list_val,
model_cdrbnr_complex_category_list_val,
model_cdrbnr_complex_value_list_val,
model_cdrbnr_complex_dataset_list_val),
columns=['CNN.Model.Name',
'Model.Metric',
'Image.Category',
'Metric.Value',
'Data.Set'])
1.6.7 Model Selection ¶
- The Simple CNN Model With No Regularization demonstrated the following validation set performance for all image categories:
- Precision = 0.7709
- Recall = 0.7576
- F1 Score = 0.7611
- The Complex CNN Model With No Regularization demonstrated the following validation set performance for all image categories:
- Precision = 0.7878
- Recall = 0.7897
- F1 Score = 0.7881
- The Simple CNN Model With Dropout Regularization demonstrated the following validation set performance for all image categories:
- Precision = 0.7709
- Recall = 0.7576
- F1 Score = 0.7611
- The Complex CNN Model With Dropout Regularization demonstrated the following validation set performance for all image categories:
- Precision = 0.7878
- Recall = 0.7897
- F1 Score = 0.7881
- The Simple CNN Model With Batch Normalization Regularization demonstrated the following validation set performance for all image categories:
- Precision = 0.8326
- Recall = 0.8261
- F1 Score = 0.8285
- The Complex CNN Model With Batch Normalization Regularization demonstrated the following validation set performance for all image categories:
- Precision = 0.6667
- Recall = 0.6539
- F1 Score = 0.6482
- The Simple CNN Model With Dropout and Batch Normalization Regularization demonstrated the following validation set performance for all image categories:
- Precision = 0.6003
- Recall = 0.4639
- F1 Score = 0.4181
- The Complex CNN Model With Dropout and Batch Normalization Regularization demonstrated the following validation set performance for all image categories:
- Precision = 0.8429
- Recall = 0.8385
- F1 Score = 0.8388
- While the classification results have been sufficiently high, the current study can be further extended to achieve optimal model performance through the following:
- Leverage pre-trained models that have been trained on large datasets to improve performance
- Conduct model hyperparameter tuning given sufficient analysis time and higher computing power
- Formulate deeper neural network architectures to better capture spatial hierarchies and features in the input images
- Apply various techniques to interpret the CNN models by understanding and visualizing the features and decisions made at each layer
- Consider an imbalanced dataset and apply remedial measures to address unbalanced classification to accurately reflect real-world scenario
In [192]:
##################################
# Consolidating all the
# CNN model performance measures
##################################
cnn_model_performance_comparison_val = pd.concat([model_nr_simple_all_summary_val,
model_nr_complex_all_summary_val,
model_dr_simple_all_summary_val,
model_dr_complex_all_summary_val,
model_bnr_simple_all_summary_val,
model_bnr_complex_all_summary_val,
model_cdrbnr_simple_all_summary_val,
model_cdrbnr_complex_all_summary_val],
ignore_index=True)
In [193]:
##################################
# Consolidating all the precision
# model performance measures
##################################
cnn_model_performance_comparison_val_precision = cnn_model_performance_comparison_val[cnn_model_performance_comparison_val['Model.Metric']=='Precision']
cnn_model_performance_comparison_val_precision_CNN_NR_Simple = cnn_model_performance_comparison_val_precision[cnn_model_performance_comparison_val_precision['CNN.Model.Name']=='CNN_NR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_precision_CNN_NR_Complex = cnn_model_performance_comparison_val_precision[cnn_model_performance_comparison_val_precision['CNN.Model.Name']=='CNN_NR_Complex'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_precision_CNN_DR_Simple = cnn_model_performance_comparison_val_precision[cnn_model_performance_comparison_val_precision['CNN.Model.Name']=='CNN_DR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_precision_CNN_DR_Complex = cnn_model_performance_comparison_val_precision[cnn_model_performance_comparison_val_precision['CNN.Model.Name']=='CNN_DR_Complex'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_precision_CNN_BNR_Simple = cnn_model_performance_comparison_val_precision[cnn_model_performance_comparison_val_precision['CNN.Model.Name']=='CNN_BNR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_precision_CNN_BNR_Complex = cnn_model_performance_comparison_val_precision[cnn_model_performance_comparison_val_precision['CNN.Model.Name']=='CNN_BNR_Complex'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_precision_CNN_CDRBNR_Simple = cnn_model_performance_comparison_val_precision[cnn_model_performance_comparison_val_precision['CNN.Model.Name']=='CNN_CDRBNR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_precision_CNN_CDRBNR_Complex = cnn_model_performance_comparison_val_precision[cnn_model_performance_comparison_val_precision['CNN.Model.Name']=='CNN_CDRBNR_Complex'].loc[:,"Metric.Value"]
In [194]:
##################################
# Combining all the precision
# model performance measures
# for all CNN models
##################################
cnn_model_performance_comparison_val_precision_plot = pd.DataFrame({'CNN_NR_Simple': cnn_model_performance_comparison_val_precision_CNN_NR_Simple.values,
'CNN_NR_Complex': cnn_model_performance_comparison_val_precision_CNN_NR_Complex.values,
'CNN_DR_Simple': cnn_model_performance_comparison_val_precision_CNN_DR_Simple.values,
'CNN_DR_Complex': cnn_model_performance_comparison_val_precision_CNN_DR_Complex.values,
'CNN_BNR_Simple': cnn_model_performance_comparison_val_precision_CNN_BNR_Simple.values,
'CNN_BNR_Complex': cnn_model_performance_comparison_val_precision_CNN_BNR_Complex.values,
'CNN_CDRBNR_Simple': cnn_model_performance_comparison_val_precision_CNN_CDRBNR_Simple.values,
'CNN_CDRBNR_Complex': cnn_model_performance_comparison_val_precision_CNN_CDRBNR_Complex.values},
index=cnn_model_performance_comparison_val_precision['Image.Category'].unique())
cnn_model_performance_comparison_val_precision_plot
Out[194]:
| CNN_NR_Simple | CNN_NR_Complex | CNN_DR_Simple | CNN_DR_Complex | CNN_BNR_Simple | CNN_BNR_Complex | CNN_CDRBNR_Simple | CNN_CDRBNR_Complex | |
|---|---|---|---|---|---|---|---|---|
| No Tumor | 0.893238 | 0.863057 | 0.852459 | 0.835913 | 0.832317 | 0.833333 | 0.722656 | 0.848765 |
| Glioma | 0.928571 | 0.871486 | 0.902778 | 0.858238 | 0.962025 | 0.623684 | 0.875000 | 0.923077 |
| Meningioma | 0.624573 | 0.655602 | 0.554054 | 0.641975 | 0.690647 | 0.430070 | 0.313199 | 0.753968 |
| Pituitary | 0.772595 | 0.795252 | 0.774691 | 0.815287 | 0.845638 | 0.780083 | 0.490698 | 0.845921 |
| Total | 0.804744 | 0.796349 | 0.770996 | 0.787853 | 0.832657 | 0.666793 | 0.600388 | 0.842933 |
In [195]:
##################################
# Plotting all the precision
# model performance measures
# for all CNN models
##################################
cnn_model_performance_comparison_val_precision_plot = cnn_model_performance_comparison_val_precision_plot.plot.barh(figsize=(10, 12), width=0.90)
cnn_model_performance_comparison_val_precision_plot.set_xlim(-0.02,1.10)
cnn_model_performance_comparison_val_precision_plot.set_title("Model Comparison by Precision Performance on Validation Data")
cnn_model_performance_comparison_val_precision_plot.set_xlabel("Precision Performance")
cnn_model_performance_comparison_val_precision_plot.set_ylabel("Image Categories")
cnn_model_performance_comparison_val_precision_plot.grid(False)
cnn_model_performance_comparison_val_precision_plot.legend(loc='center left', bbox_to_anchor=(1.0, 0.5))
for container in cnn_model_performance_comparison_val_precision_plot.containers:
cnn_model_performance_comparison_val_precision_plot.bar_label(container, fmt='%.5f', padding=+10, color='black', fontweight='bold')
In [196]:
##################################
# Consolidating all the recall
# model performance measures
##################################
cnn_model_performance_comparison_val_recall = cnn_model_performance_comparison_val[cnn_model_performance_comparison_val['Model.Metric']=='Recall']
cnn_model_performance_comparison_val_recall_CNN_NR_Simple = cnn_model_performance_comparison_val_recall[cnn_model_performance_comparison_val_recall['CNN.Model.Name']=='CNN_NR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_recall_CNN_NR_Complex = cnn_model_performance_comparison_val_recall[cnn_model_performance_comparison_val_recall['CNN.Model.Name']=='CNN_NR_Complex'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_recall_CNN_DR_Simple = cnn_model_performance_comparison_val_recall[cnn_model_performance_comparison_val_recall['CNN.Model.Name']=='CNN_DR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_recall_CNN_DR_Complex = cnn_model_performance_comparison_val_recall[cnn_model_performance_comparison_val_recall['CNN.Model.Name']=='CNN_DR_Complex'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_recall_CNN_BNR_Simple = cnn_model_performance_comparison_val_recall[cnn_model_performance_comparison_val_recall['CNN.Model.Name']=='CNN_BNR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_recall_CNN_BNR_Complex = cnn_model_performance_comparison_val_recall[cnn_model_performance_comparison_val_recall['CNN.Model.Name']=='CNN_BNR_Complex'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_recall_CNN_CDRBNR_Simple = cnn_model_performance_comparison_val_recall[cnn_model_performance_comparison_val_recall['CNN.Model.Name']=='CNN_CDRBNR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_recall_CNN_CDRBNR_Complex = cnn_model_performance_comparison_val_recall[cnn_model_performance_comparison_val_recall['CNN.Model.Name']=='CNN_CDRBNR_Complex'].loc[:,"Metric.Value"]
In [197]:
##################################
# Combining all the recall
# model performance measures
# for all CNN models
##################################
cnn_model_performance_comparison_val_recall_plot = pd.DataFrame({'CNN_NR_Simple': cnn_model_performance_comparison_val_recall_CNN_NR_Simple.values,
'CNN_NR_Complex': cnn_model_performance_comparison_val_recall_CNN_NR_Complex.values,
'CNN_DR_Simple': cnn_model_performance_comparison_val_recall_CNN_DR_Simple.values,
'CNN_DR_Complex': cnn_model_performance_comparison_val_recall_CNN_DR_Complex.values,
'CNN_BNR_Simple': cnn_model_performance_comparison_val_recall_CNN_BNR_Simple.values,
'CNN_BNR_Complex': cnn_model_performance_comparison_val_recall_CNN_BNR_Complex.values,
'CNN_CDRBNR_Simple': cnn_model_performance_comparison_val_recall_CNN_CDRBNR_Simple.values,
'CNN_CDRBNR_Complex': cnn_model_performance_comparison_val_recall_CNN_CDRBNR_Complex.values},
index=cnn_model_performance_comparison_val_recall['Image.Category'].unique())
cnn_model_performance_comparison_val_recall_plot
Out[197]:
| CNN_NR_Simple | CNN_NR_Complex | CNN_DR_Simple | CNN_DR_Complex | CNN_BNR_Simple | CNN_BNR_Complex | CNN_CDRBNR_Simple | CNN_CDRBNR_Complex | |
|---|---|---|---|---|---|---|---|---|
| No Tumor | 0.786834 | 0.849530 | 0.815047 | 0.846395 | 0.855799 | 0.611285 | 0.579937 | 0.862069 |
| Glioma | 0.787879 | 0.821970 | 0.738636 | 0.848485 | 0.863636 | 0.897727 | 0.026515 | 0.818182 |
| Meningioma | 0.685393 | 0.591760 | 0.614232 | 0.584270 | 0.719101 | 0.460674 | 0.524345 | 0.711610 |
| Pituitary | 0.910653 | 0.920962 | 0.862543 | 0.879725 | 0.865979 | 0.646048 | 0.725086 | 0.962199 |
| Total | 0.792690 | 0.796055 | 0.757615 | 0.789719 | 0.826129 | 0.653934 | 0.463971 | 0.838515 |
In [198]:
##################################
# Plotting all the recall
# model performance measures
# for all CNN models
##################################
cnn_model_performance_comparison_val_recall_plot = cnn_model_performance_comparison_val_recall_plot.plot.barh(figsize=(10, 12), width=0.90)
cnn_model_performance_comparison_val_recall_plot.set_xlim(-0.02,1.10)
cnn_model_performance_comparison_val_recall_plot.set_title("Model Comparison by Recall Performance on Validation Data")
cnn_model_performance_comparison_val_recall_plot.set_xlabel("Recall Performance")
cnn_model_performance_comparison_val_recall_plot.set_ylabel("Image Categories")
cnn_model_performance_comparison_val_recall_plot.grid(False)
cnn_model_performance_comparison_val_recall_plot.legend(loc='center left', bbox_to_anchor=(1.0, 0.5))
for container in cnn_model_performance_comparison_val_recall_plot.containers:
cnn_model_performance_comparison_val_recall_plot.bar_label(container, fmt='%.5f', padding=+10, color='black', fontweight='bold')
In [199]:
##################################
# Consolidating all the fscore
# model performance measures
##################################
cnn_model_performance_comparison_val_fscore = cnn_model_performance_comparison_val[cnn_model_performance_comparison_val['Model.Metric']=='F-Score']
cnn_model_performance_comparison_val_fscore_CNN_NR_Simple = cnn_model_performance_comparison_val_fscore[cnn_model_performance_comparison_val_fscore['CNN.Model.Name']=='CNN_NR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_fscore_CNN_NR_Complex = cnn_model_performance_comparison_val_fscore[cnn_model_performance_comparison_val_fscore['CNN.Model.Name']=='CNN_NR_Complex'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_fscore_CNN_DR_Simple = cnn_model_performance_comparison_val_fscore[cnn_model_performance_comparison_val_fscore['CNN.Model.Name']=='CNN_DR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_fscore_CNN_DR_Complex = cnn_model_performance_comparison_val_fscore[cnn_model_performance_comparison_val_fscore['CNN.Model.Name']=='CNN_DR_Complex'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_fscore_CNN_BNR_Simple = cnn_model_performance_comparison_val_fscore[cnn_model_performance_comparison_val_fscore['CNN.Model.Name']=='CNN_BNR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_fscore_CNN_BNR_Complex = cnn_model_performance_comparison_val_fscore[cnn_model_performance_comparison_val_fscore['CNN.Model.Name']=='CNN_BNR_Complex'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_fscore_CNN_CDRBNR_Simple = cnn_model_performance_comparison_val_fscore[cnn_model_performance_comparison_val_fscore['CNN.Model.Name']=='CNN_CDRBNR_Simple'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_fscore_CNN_CDRBNR_Complex = cnn_model_performance_comparison_val_fscore[cnn_model_performance_comparison_val_fscore['CNN.Model.Name']=='CNN_CDRBNR_Complex'].loc[:,"Metric.Value"]
In [200]:
##################################
# Combining all the fscore
# model performance measures
# for all CNN models
##################################
cnn_model_performance_comparison_val_fscore_plot = pd.DataFrame({'CNN_NR_Simple': cnn_model_performance_comparison_val_fscore_CNN_NR_Simple.values,
'CNN_NR_Complex': cnn_model_performance_comparison_val_fscore_CNN_NR_Complex.values,
'CNN_DR_Simple': cnn_model_performance_comparison_val_fscore_CNN_DR_Simple.values,
'CNN_DR_Complex': cnn_model_performance_comparison_val_fscore_CNN_DR_Complex.values,
'CNN_BNR_Simple': cnn_model_performance_comparison_val_fscore_CNN_BNR_Simple.values,
'CNN_BNR_Complex': cnn_model_performance_comparison_val_fscore_CNN_BNR_Complex.values,
'CNN_CDRBNR_Simple': cnn_model_performance_comparison_val_fscore_CNN_CDRBNR_Simple.values,
'CNN_CDRBNR_Complex': cnn_model_performance_comparison_val_fscore_CNN_CDRBNR_Complex.values},
index=cnn_model_performance_comparison_val_fscore['Image.Category'].unique())
cnn_model_performance_comparison_val_fscore_plot
Out[200]:
| CNN_NR_Simple | CNN_NR_Complex | CNN_DR_Simple | CNN_DR_Complex | CNN_BNR_Simple | CNN_BNR_Complex | CNN_CDRBNR_Simple | CNN_CDRBNR_Complex | |
|---|---|---|---|---|---|---|---|---|
| No Tumor | 0.836667 | 0.856240 | 0.833333 | 0.841121 | 0.843895 | 0.705244 | 0.643478 | 0.855365 |
| Glioma | 0.852459 | 0.846004 | 0.812500 | 0.853333 | 0.910180 | 0.736025 | 0.051471 | 0.867470 |
| Meningioma | 0.653571 | 0.622047 | 0.582593 | 0.611765 | 0.704587 | 0.444846 | 0.392157 | 0.732177 |
| Pituitary | 0.835962 | 0.853503 | 0.816260 | 0.846281 | 0.855688 | 0.706767 | 0.585298 | 0.900322 |
| Total | 0.794665 | 0.794449 | 0.761172 | 0.788125 | 0.828587 | 0.648221 | 0.418101 | 0.838834 |
In [201]:
##################################
# Plotting all the fscore
# model performance measures
# for all CNN models
##################################
cnn_model_performance_comparison_val_fscore_plot = cnn_model_performance_comparison_val_fscore_plot.plot.barh(figsize=(10, 12), width=0.90)
cnn_model_performance_comparison_val_fscore_plot.set_xlim(-0.02,1.10)
cnn_model_performance_comparison_val_fscore_plot.set_title("Model Comparison by F-Score Performance on Validation Data")
cnn_model_performance_comparison_val_fscore_plot.set_xlabel("F-Score Performance")
cnn_model_performance_comparison_val_fscore_plot.set_ylabel("Image Categories")
cnn_model_performance_comparison_val_fscore_plot.grid(False)
cnn_model_performance_comparison_val_fscore_plot.legend(loc='center left', bbox_to_anchor=(1.0, 0.5))
for container in cnn_model_performance_comparison_val_fscore_plot.containers:
cnn_model_performance_comparison_val_fscore_plot.bar_label(container, fmt='%.5f', padding=+10, color='black', fontweight='bold')
1.6.8 Model Testing ¶
In [202]:
##################################
# Evaluating the model
# for a complex CNN with dropout and batch normalization regularization
# on the independent test set
##################################
model_cdrbnr_complex_y_pred_test = model_cdrbnr_complex.predict(test_gen)
41/41 ━━━━━━━━━━━━━━━━━━━━ 4s 102ms/step
In [203]:
##################################
# Consolidating the predictions
# for a complex CNN with dropout and batch normalization regularization
# on the test set
##################################
model_cdrbnr_complex_predictions_test = np.array(list(map(lambda x: np.argmax(x), model_cdrbnr_complex_y_pred_test)))
model_cdrbnr_complex_y_true_test = test_gen.classes
##################################
# Formulating the confusion matrix
# for a complex CNN with dropout and batch normalization regularization
# on the test set
##################################
cmatrix_test = pd.DataFrame(confusion_matrix(model_cdrbnr_complex_y_true_test, model_cdrbnr_complex_predictions_test), columns=classes, index =classes)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with dropout and batch normalization regularization
# for each category of the test set
##################################
plt.figure(figsize=(10, 6))
ax = sns.heatmap(cmatrix_test, annot = True, fmt = 'g' ,vmin = 0, vmax = 250, cmap = 'icefire')
ax.set_xlabel('Predicted',fontsize = 14,weight = 'bold')
ax.set_xticklabels(ax.get_xticklabels(),rotation =0)
ax.set_ylabel('Actual',fontsize = 14,weight = 'bold')
ax.set_yticklabels(ax.get_yticklabels(),rotation =0)
ax.set_title('Complex CNN With Dropout and Batch Normalization Regularization : Test Set Confusion Matrix',fontsize = 14, weight = 'bold', pad=20);
In [204]:
##################################
# Calculating the model accuracy
# for a complex CNN with dropout and batch normalization regularization
# for the entire test set
##################################
model_cdrbnr_complex_acc_test = accuracy_score(model_cdrbnr_complex_y_true_test, model_cdrbnr_complex_predictions_test)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with dropout and batch normalization regularization
# for the entire test set
##################################
model_cdrbnr_complex_results_all_test = precision_recall_fscore_support(model_cdrbnr_complex_y_true_test, model_cdrbnr_complex_predictions_test, average='macro', zero_division = 1)
##################################
# Calculating the model
# Precision, Recall, F-score and Support
# for a complex CNN with dropout and batch normalization regularization
# for each category of the test set
##################################
model_cdrbnr_complex_results_class_test = precision_recall_fscore_support(model_cdrbnr_complex_y_true_test, model_cdrbnr_complex_predictions_test, average=None, zero_division = 1)
##################################
# Consolidating all model evaluation metrics
# for a complex CNN with dropout and batch normalization regularization
##################################
metric_columns = ['Precision','Recall', 'F-Score','Support']
model_cdrbnr_complex_all_df_test = pd.concat([pd.DataFrame(list(model_cdrbnr_complex_results_class_test)).T,pd.DataFrame(list(model_cdrbnr_complex_results_all_test)).T])
model_cdrbnr_complex_all_df_test.columns = metric_columns
model_cdrbnr_complex_all_df_test.index = ['No Tumor', 'Glioma', 'Meningioma', 'Pituitary', 'Total']
print('Complex CNN With Dropout and Batch Normalization Regularization : Test Set Classification Performance')
model_cdrbnr_complex_all_df_test
Complex CNN With Dropout and Batch Normalization Regularization : Test Set Classification Performance
Out[204]:
| Precision | Recall | F-Score | Support | |
|---|---|---|---|---|
| No Tumor | 0.860215 | 0.987654 | 0.919540 | 405.0 |
| Glioma | 0.923345 | 0.883333 | 0.902896 | 300.0 |
| Meningioma | 0.864542 | 0.709150 | 0.779174 | 306.0 |
| Pituitary | 0.938312 | 0.963333 | 0.950658 | 300.0 |
| Total | 0.896603 | 0.885868 | 0.888067 | NaN |
In [205]:
##################################
# Consolidating all model evaluation metrics
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
model_cdrbnr_complex_model_list_test = []
model_cdrbnr_complex_measure_list_test = []
model_cdrbnr_complex_category_list_test = []
model_cdrbnr_complex_value_list_test = []
model_cdrbnr_complex_dataset_list_test = []
for i in range(3):
for j in range(5):
model_cdrbnr_complex_model_list_test.append('CNN_CDRBNR_Complex')
model_cdrbnr_complex_measure_list_test.append(metric_columns[i])
model_cdrbnr_complex_category_list_test.append(model_cdrbnr_complex_all_df_test.index[j])
model_cdrbnr_complex_value_list_test.append(model_cdrbnr_complex_all_df_test.iloc[j,i])
model_cdrbnr_complex_dataset_list_test.append('Test')
model_cdrbnr_complex_all_summary_test = pd.DataFrame(zip(model_cdrbnr_complex_model_list_test,
model_cdrbnr_complex_measure_list_test,
model_cdrbnr_complex_category_list_test,
model_cdrbnr_complex_value_list_test,
model_cdrbnr_complex_dataset_list_test),
columns=['CNN.Model.Name',
'Model.Metric',
'Image.Category',
'Metric.Value',
'Data.Set'])
In [206]:
##################################
# Consolidating all the
# CNN model performance measures
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test = pd.concat([model_cdrbnr_complex_all_summary_val,
model_cdrbnr_complex_all_summary_test],
ignore_index=True)
In [207]:
##################################
# Consolidating all the precision
# model performance measures
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test_precision = cnn_model_performance_comparison_val_test[cnn_model_performance_comparison_val_test['Model.Metric']=='Precision']
cnn_model_performance_comparison_val_test_precision_CNN_CDRBNR_Complex_validation = cnn_model_performance_comparison_val_test_precision[cnn_model_performance_comparison_val_test_precision['Data.Set']=='Validation'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_test_precision_CNN_CDRBNR_Complex_test = cnn_model_performance_comparison_val_test_precision[cnn_model_performance_comparison_val_test_precision['Data.Set']=='Test'].loc[:,"Metric.Value"]
In [208]:
##################################
# Combining all the precision
# model performance measures
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test_precision_plot = pd.DataFrame({'CNN_CDRBNR_Complex_Validation': cnn_model_performance_comparison_val_test_precision_CNN_CDRBNR_Complex_validation.values,
'CNN_CDRBNR_Complex_Test': cnn_model_performance_comparison_val_test_precision_CNN_CDRBNR_Complex_test.values},
cnn_model_performance_comparison_val_test_precision['Image.Category'].unique())
cnn_model_performance_comparison_val_test_precision_plot
Out[208]:
| CNN_CDRBNR_Complex_Validation | CNN_CDRBNR_Complex_Test | |
|---|---|---|
| No Tumor | 0.848765 | 0.860215 |
| Glioma | 0.923077 | 0.923345 |
| Meningioma | 0.753968 | 0.864542 |
| Pituitary | 0.845921 | 0.938312 |
| Total | 0.842933 | 0.896603 |
In [209]:
##################################
# Plotting all the precision
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test_precision_plot = cnn_model_performance_comparison_val_test_precision_plot.plot.barh(figsize=(10, 6), width=0.90)
cnn_model_performance_comparison_val_test_precision_plot.set_xlim(-0.02,1.10)
cnn_model_performance_comparison_val_test_precision_plot.set_title("Model Precision Performance Comparison on Validation and Test Data")
cnn_model_performance_comparison_val_test_precision_plot.set_xlabel("Precision Performance")
cnn_model_performance_comparison_val_test_precision_plot.set_ylabel("Image Categories")
cnn_model_performance_comparison_val_test_precision_plot.grid(False)
cnn_model_performance_comparison_val_test_precision_plot.legend(loc='center left', bbox_to_anchor=(1.0, 0.5))
for container in cnn_model_performance_comparison_val_test_precision_plot.containers:
cnn_model_performance_comparison_val_test_precision_plot.bar_label(container, fmt='%.5f', padding=-50, color='white', fontweight='bold')
In [210]:
##################################
# Consolidating all the recall
# model performance measures
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test_recall = cnn_model_performance_comparison_val_test[cnn_model_performance_comparison_val_test['Model.Metric']=='Recall']
cnn_model_performance_comparison_val_test_recall_CNN_CDRBNR_Complex_validation = cnn_model_performance_comparison_val_test_recall[cnn_model_performance_comparison_val_test_recall['Data.Set']=='Validation'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_test_recall_CNN_CDRBNR_Complex_test = cnn_model_performance_comparison_val_test_recall[cnn_model_performance_comparison_val_test_recall['Data.Set']=='Test'].loc[:,"Metric.Value"]
In [211]:
##################################
# Combining all the recall
# model performance measures
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test_recall_plot = pd.DataFrame({'CNN_CDRBNR_Complex_Validation': cnn_model_performance_comparison_val_test_recall_CNN_CDRBNR_Complex_validation.values,
'CNN_CDRBNR_Complex_Test': cnn_model_performance_comparison_val_test_recall_CNN_CDRBNR_Complex_test.values},
cnn_model_performance_comparison_val_test_recall['Image.Category'].unique())
cnn_model_performance_comparison_val_test_recall_plot
Out[211]:
| CNN_CDRBNR_Complex_Validation | CNN_CDRBNR_Complex_Test | |
|---|---|---|
| No Tumor | 0.862069 | 0.987654 |
| Glioma | 0.818182 | 0.883333 |
| Meningioma | 0.711610 | 0.709150 |
| Pituitary | 0.962199 | 0.963333 |
| Total | 0.838515 | 0.885868 |
In [212]:
##################################
# Plotting all the recall
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test_recall_plot = cnn_model_performance_comparison_val_test_recall_plot.plot.barh(figsize=(10, 6), width=0.90)
cnn_model_performance_comparison_val_test_recall_plot.set_xlim(-0.02,1.10)
cnn_model_performance_comparison_val_test_recall_plot.set_title("Model Recall Performance Comparison on Validation and Test Data")
cnn_model_performance_comparison_val_test_recall_plot.set_xlabel("Recall Performance")
cnn_model_performance_comparison_val_test_recall_plot.set_ylabel("Image Categories")
cnn_model_performance_comparison_val_test_recall_plot.grid(False)
cnn_model_performance_comparison_val_test_recall_plot.legend(loc='center left', bbox_to_anchor=(1.0, 0.5))
for container in cnn_model_performance_comparison_val_test_recall_plot.containers:
cnn_model_performance_comparison_val_test_recall_plot.bar_label(container, fmt='%.5f', padding=-50, color='white', fontweight='bold')
In [213]:
##################################
# Consolidating all the fscore
# model performance measures
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test_fscore = cnn_model_performance_comparison_val_test[cnn_model_performance_comparison_val_test['Model.Metric']=='F-Score']
cnn_model_performance_comparison_val_test_fscore_CNN_CDRBNR_Complex_validation = cnn_model_performance_comparison_val_test_fscore[cnn_model_performance_comparison_val_test_fscore['Data.Set']=='Validation'].loc[:,"Metric.Value"]
cnn_model_performance_comparison_val_test_fscore_CNN_CDRBNR_Complex_test = cnn_model_performance_comparison_val_test_fscore[cnn_model_performance_comparison_val_test_fscore['Data.Set']=='Test'].loc[:,"Metric.Value"]
In [214]:
##################################
# Combining all the fscore
# model performance measures
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test_fscore_plot = pd.DataFrame({'CNN_CDRBNR_Complex_Validation': cnn_model_performance_comparison_val_test_fscore_CNN_CDRBNR_Complex_validation.values,
'CNN_CDRBNR_Complex_Test': cnn_model_performance_comparison_val_test_fscore_CNN_CDRBNR_Complex_test.values},
cnn_model_performance_comparison_val_test_fscore['Image.Category'].unique())
cnn_model_performance_comparison_val_test_fscore_plot
Out[214]:
| CNN_CDRBNR_Complex_Validation | CNN_CDRBNR_Complex_Test | |
|---|---|---|
| No Tumor | 0.855365 | 0.919540 |
| Glioma | 0.867470 | 0.902896 |
| Meningioma | 0.732177 | 0.779174 |
| Pituitary | 0.900322 | 0.950658 |
| Total | 0.838834 | 0.888067 |
In [215]:
##################################
# Plotting all the fscore
# for the selected model defined as
# complex CNN with dropout and batch normalization regularization
##################################
cnn_model_performance_comparison_val_test_fscore_plot = cnn_model_performance_comparison_val_test_fscore_plot.plot.barh(figsize=(10, 6), width=0.90)
cnn_model_performance_comparison_val_test_fscore_plot.set_xlim(-0.02,1.10)
cnn_model_performance_comparison_val_test_fscore_plot.set_title("Model F-Score Performance Comparison on Validation and Test Data")
cnn_model_performance_comparison_val_test_fscore_plot.set_xlabel("F-Score Performance")
cnn_model_performance_comparison_val_test_fscore_plot.set_ylabel("Image Categories")
cnn_model_performance_comparison_val_test_fscore_plot.grid(False)
cnn_model_performance_comparison_val_test_fscore_plot.legend(loc='center left', bbox_to_anchor=(1.0, 0.5))
for container in cnn_model_performance_comparison_val_test_fscore_plot.containers:
cnn_model_performance_comparison_val_test_fscore_plot.bar_label(container, fmt='%.5f', padding=-50, color='white', fontweight='bold')
1.6.9 Model Inference ¶
In [216]:
##################################
# Gathering the actual and predicted classes
# from the selected CNN model defined as
# complex CNN with dropout and batch normalization regularization
##################################
model_cdrbnr_complex_predictions_test = np.array(list(map(lambda x: np.argmax(x), model_cdrbnr_complex_y_pred_test)))
model_cdrbnr_complex_y_true_test = test_gen.classes
In [217]:
##################################
# Consolidating the actual and predicted classes
# from the selected CNN model defined as
# complex CNN with dropout and batch normalization regularization
##################################
class_indices = test_gen.class_indices
indices = {v:k for k,v in class_indices.items()}
filenames = test_gen.filenames
test_gen_df = pd.DataFrame()
test_gen_df['FileName'] = filenames
test_gen_df['Actual_Category'] = model_cdrbnr_complex_y_true_test
test_gen_df['Predicted_Category'] = model_cdrbnr_complex_predictions_test
test_gen_df['Actual_Category'] = test_gen_df['Actual_Category'].apply(lambda x: indices[x])
test_gen_df['Predicted_Category'] = test_gen_df['Predicted_Category'].apply(lambda x: indices[x])
test_gen_df.loc[test_gen_df['Actual_Category']==test_gen_df['Predicted_Category'],'Matched_Category_Prediction'] = True
test_gen_df.loc[test_gen_df['Actual_Category']!=test_gen_df['Predicted_Category'],'Matched_Category_Prediction'] = False
test_gen_df.head(10)
Out[217]:
| FileName | Actual_Category | Predicted_Category | Matched_Category_Prediction | |
|---|---|---|---|---|
| 0 | notumor\Te-noTr_0000.jpg | notumor | notumor | True |
| 1 | notumor\Te-noTr_0001.jpg | notumor | notumor | True |
| 2 | notumor\Te-noTr_0002.jpg | notumor | notumor | True |
| 3 | notumor\Te-noTr_0003.jpg | notumor | notumor | True |
| 4 | notumor\Te-noTr_0004.jpg | notumor | notumor | True |
| 5 | notumor\Te-noTr_0005.jpg | notumor | notumor | True |
| 6 | notumor\Te-noTr_0006.jpg | notumor | notumor | True |
| 7 | notumor\Te-noTr_0007.jpg | notumor | notumor | True |
| 8 | notumor\Te-noTr_0008.jpg | notumor | notumor | True |
| 9 | notumor\Te-noTr_0009.jpg | notumor | notumor | True |
In [218]:
##################################
# Formulating image samples
# from the validation set
##################################
test_gen_df = test_gen_df.sample(frac=1, replace=False, random_state=123).reset_index(drop=True)
In [219]:
##################################
# Defining a function
# to load the sampled images
##################################
img_size=227
def readImage(path):
img = load_img(path,color_mode="grayscale", target_size=(img_size,img_size))
img = img_to_array(img)
img = img/255.
return img
In [220]:
##################################
# Defining a function
# to display the sampled images
# with the actual and predicted categories
##################################
base_path = (os.path.join("..", DATASETS_FINAL_TEST_PATH))
def display_images(temp_df):
temp_df = temp_df.reset_index(drop=True)
plt.figure(figsize = (20 , 20))
n = 0
for i in range(15):
n+=1
plt.subplot(5 , 5, n)
plt.subplots_adjust(hspace = 0.5 , wspace = 0.3)
image = readImage(f"{base_path}\\{temp_df.FileName[i]}")
plt.imshow(image)
plt.title(f'A: {temp_df.Actual_Category[i]} P: {temp_df.Predicted_Category[i]}')
In [221]:
##################################
# Display sample images with matched
# actual and predicted categories
##################################
display_images(test_gen_df[test_gen_df['Matched_Category_Prediction']==True])
In [222]:
##################################
# Display sample images with mismatched
# actual and predicted categories
##################################
display_images(test_gen_df[test_gen_df['Matched_Category_Prediction']!=True])
In [223]:
##################################
# Recreating the CNN model defined as
# complex CNN with dropout and batch normalization regularization
# using the Functional API structure
##################################
##################################
# Defining the input layer
##################################
fmodel_input_layer = Input(shape=(227, 227, 1), name="input_layer")
##################################
# Using the layers from the Sequential model
# as functions in the Functional API
##################################
set_seed()
fmodel_conv2d_layer = model_cdrbnr_complex.layers[0](fmodel_input_layer) # Conv2D layer
fmodel_maxpooling2d_layer = model_cdrbnr_complex.layers[1](fmodel_conv2d_layer) # MaxPooling2D layer
fmodel_conv2d_1_layer = model_cdrbnr_complex.layers[2](fmodel_maxpooling2d_layer) # Conv2D layer
fmodel_maxpooling2d_1_layer = model_cdrbnr_complex.layers[3](fmodel_conv2d_1_layer) # MaxPooling2D layer
fmodel_conv2d_2_layer = model_cdrbnr_complex.layers[4](fmodel_maxpooling2d_1_layer) # Conv2D layer
fmodel_batchnormalization_layer = model_cdrbnr_complex.layers[5](fmodel_conv2d_2_layer) # Batch Normalization layer
fmodel_activation_layer = model_cdrbnr_complex.layers[6](fmodel_batchnormalization_layer) # Activation layer
fmodel_maxpooling2d_2_layer = model_cdrbnr_complex.layers[7](fmodel_activation_layer) # MaxPooling2D layer
fmodel_flatten_layer = model_cdrbnr_complex.layers[8](fmodel_maxpooling2d_2_layer) # Flatten layer
fmodel_dense_layer = model_cdrbnr_complex.layers[9](fmodel_flatten_layer) # Dense layer (128 units)
fmodel_dropout_layer = model_cdrbnr_complex.layers[10](fmodel_dense_layer) # Dropout layer
fmodel_output_layer = model_cdrbnr_complex.layers[11](fmodel_dropout_layer) # Dense layer (num_classes units)
##################################
# Creating the Functional API model
##################################
model_cdrbnr_complex_functional_api = Model(inputs=fmodel_input_layer, outputs=fmodel_output_layer, name="model_cdrbnr_complex_fapi")
##################################
# Compiling the Functional API model
# with the same parameters
##################################
set_seed()
model_cdrbnr_complex_functional_api.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
##################################
# Displaying the model summary
# for CNN with dropout regularization
##################################
print(model_cdrbnr_complex_functional_api.summary())
Model: "model_cdrbnr_complex_fapi"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ input_layer (InputLayer) │ (None, 227, 227, 1) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_conv2d_0 (Conv2D) │ (None, 227, 227, 16) │ 160 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_max_pooling2d_0 │ (None, 113, 113, 16) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_conv2d_1 (Conv2D) │ (None, 113, 113, 32) │ 4,640 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_max_pooling2d_1 │ (None, 56, 56, 32) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_conv2d_2 (Conv2D) │ (None, 56, 56, 64) │ 18,496 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_batch_normalization │ (None, 56, 56, 64) │ 256 │ │ (BatchNormalization) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_activation │ (None, 56, 56, 64) │ 0 │ │ (Activation) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_max_pooling2d_2 │ (None, 28, 28, 64) │ 0 │ │ (MaxPooling2D) │ │ │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_flatten (Flatten) │ (None, 50176) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_dense_0 (Dense) │ (None, 128) │ 6,422,656 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_dropout (Dropout) │ (None, 128) │ 0 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ cdrbnr_complex_dense_1 (Dense) │ (None, 4) │ 516 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 6,446,724 (24.59 MB)
Trainable params: 6,446,596 (24.59 MB)
Non-trainable params: 128 (512.00 B)
None
In [224]:
##################################
# Creating a gradient model for the
# gradient class activation map
# of the first convolutional layer
##################################
grad_model_first_conv2d = Model(inputs=fmodel_input_layer, outputs=[fmodel_conv2d_layer, fmodel_output_layer], name="model_cdrbnr_complex_fapi_first_conv2d")
set_seed()
grad_model_first_conv2d.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [225]:
##################################
# Defining a function
# to formulate the gradient class activation map
# from the output of the first convolutional layer
##################################
def make_gradcam_heatmap(img_array, pred_index=None):
with tf.GradientTape() as tape:
last_conv_layer_output, preds = grad_model_first_conv2d(img_array)
if pred_index is None:
pred_index = tf.argmax(preds[0])
class_channel = preds[:, pred_index]
grads = tape.gradient(class_channel, last_conv_layer_output)
pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))
last_conv_layer_output = last_conv_layer_output[0]
heatmap = last_conv_layer_output @ pooled_grads[..., tf.newaxis]
heatmap = tf.squeeze(heatmap)
heatmap = tf.maximum(heatmap, 0) / tf.math.reduce_max(heatmap)
return heatmap.numpy(), preds
In [226]:
##################################
# Defining a function
# to colorize the generated heatmap
# and superimpose on the actual image
##################################
def gradCAMImage(image):
path = (os.path.join("..", DATASETS_FINAL_TEST_PATH, image))
img = readImage(path)
img = np.expand_dims(img,axis=0)
heatmap, preds = make_gradcam_heatmap(img)
img = load_img(path)
img = img_to_array(img)
heatmap = np.uint8(255 * heatmap)
jet = plt.colormaps["turbo"]
jet_colors = jet(np.arange(256))[:, :3]
jet_heatmap = jet_colors[heatmap]
jet_heatmap = tf.keras.preprocessing.image.array_to_img(jet_heatmap)
jet_heatmap = jet_heatmap.resize((img.shape[1], img.shape[0]))
jet_heatmap = tf.keras.preprocessing.image.img_to_array(jet_heatmap)
superimposed_img = jet_heatmap * 0.80 + img
superimposed_img = tf.keras.preprocessing.image.array_to_img(superimposed_img)
return superimposed_img
In [227]:
##################################
# Defining a function to consolidate
# the gradient class activation maps
# for a subset of sampled images
##################################
def gradcam_of_images(correct_class):
grad_images = []
title = []
temp_df = test_gen_df[test_gen_df['Matched_Category_Prediction']==correct_class]
temp_df = temp_df.reset_index(drop=True)
for i in range(15):
image = temp_df.FileName[i]
grad_image = gradCAMImage(image)
grad_images.append(grad_image)
title.append(f"A: {temp_df.Actual_Category[i]} P: {temp_df.Predicted_Category[i]}")
return grad_images, title
In [228]:
##################################
# Consolidating the gradient class activation maps
# from the output of the first convolutional layer
# for the subset of sampled images
# with matched actual and predicted categories
##################################
matched_categories, matched_categories_titles = gradcam_of_images(correct_class=True)
In [229]:
##################################
# Consolidating the gradient class activation maps
# from the output of the first convolutional layer
# for the subset of sampled images
# with mismatched actual and predicted categories
##################################
mismatched_categories, mismatched_categories_titles = gradcam_of_images(correct_class=False)
In [230]:
##################################
# Defining a function to display
# the consolidated gradient class activation maps
# for a subset of sampled images
##################################
def display_heatmaps(classified_images, titles):
plt.figure(figsize = (20 , 20))
n = 0
for i in range(15):
n+=1
plt.subplot(5 , 5, n)
plt.subplots_adjust(hspace = 0.5 , wspace = 0.3)
plt.imshow(classified_images[i])
plt.title(titles[i])
plt.show()
In [231]:
##################################
# Displaying the consolidated
# gradient class activation maps
# from the output of the first convolutional layer
# for the subset of sampled images
# with matched actual and predicted categories
##################################
display_heatmaps(matched_categories, matched_categories_titles)
In [232]:
##################################
# Displaying the consolidated
# gradient class activation maps
# from the output of the first convolutional layer
# for the subset of sampled images
# with mismatched actual and predicted categories
##################################
display_heatmaps(mismatched_categories, mismatched_categories_titles)
In [233]:
##################################
# Creating a gradient model for the
# gradient class activation map
# of the second convolutional layer
##################################
grad_model_second_conv2d = Model(inputs=fmodel_input_layer, outputs=[fmodel_conv2d_1_layer, fmodel_output_layer], name="model_cdrbnr_complex_fapi_second_conv2d")
set_seed()
grad_model_second_conv2d.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [234]:
##################################
# Defining a function
# to formulate the gradient class activation map
# from the output of the second convolutional layer
##################################
def make_gradcam_heatmap(img_array, pred_index=None):
with tf.GradientTape() as tape:
last_conv_layer_output, preds = grad_model_second_conv2d(img_array)
if pred_index is None:
pred_index = tf.argmax(preds[0])
class_channel = preds[:, pred_index]
grads = tape.gradient(class_channel, last_conv_layer_output)
pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))
last_conv_layer_output = last_conv_layer_output[0]
heatmap = last_conv_layer_output @ pooled_grads[..., tf.newaxis]
heatmap = tf.squeeze(heatmap)
heatmap = tf.maximum(heatmap, 0) / tf.math.reduce_max(heatmap)
return heatmap.numpy(), preds
In [235]:
##################################
# Consolidating the gradient class activation maps
# from the output of the second convolutional layer
# for the subset of sampled images
# with matched actual and predicted categories
##################################
matched_categories, matched_categories_titles = gradcam_of_images(correct_class=True)
In [236]:
##################################
# Consolidating the gradient class activation maps
# from the output of the second convolutional layer
# for the subset of sampled images
# with mismatched actual and predicted categories
##################################
mismatched_categories, mismatched_categories_titles = gradcam_of_images(correct_class=False)
In [237]:
##################################
# Displaying the consolidated
# gradient class activation maps
# from the output of the second convolutional layer
# for the subset of sampled images
# with matched actual and predicted categories
##################################
display_heatmaps(matched_categories, matched_categories_titles)
In [238]:
##################################
# Displaying the consolidated
# gradient class activation maps
# from the output of the second convolutional layer
# for the subset of sampled images
# with mismatched actual and predicted categories
##################################
display_heatmaps(mismatched_categories, mismatched_categories_titles)
In [239]:
##################################
# Creating a gradient model for the
# gradient class activation map
# of the third convolutional layer
##################################
grad_model_third_conv2d = Model(inputs=fmodel_input_layer, outputs=[fmodel_conv2d_2_layer, fmodel_output_layer], name="model_cdrbnr_complex_fapi_third_conv2d")
set_seed()
grad_model_third_conv2d.compile(loss='categorical_crossentropy', optimizer='adam', metrics=[Recall(name='recall')])
In [240]:
##################################
# Defining a function
# to formulate the gradient class activation map
# from the output of the third convolutional layer
##################################
def make_gradcam_heatmap(img_array, pred_index=None):
with tf.GradientTape() as tape:
last_conv_layer_output, preds = grad_model_third_conv2d(img_array)
if pred_index is None:
pred_index = tf.argmax(preds[0])
class_channel = preds[:, pred_index]
grads = tape.gradient(class_channel, last_conv_layer_output)
pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))
last_conv_layer_output = last_conv_layer_output[0]
heatmap = last_conv_layer_output @ pooled_grads[..., tf.newaxis]
heatmap = tf.squeeze(heatmap)
heatmap = tf.maximum(heatmap, 0) / tf.math.reduce_max(heatmap)
return heatmap.numpy(), preds
In [241]:
##################################
# Consolidating the gradient class activation maps
# from the output of the third convolutional layer
# for the subset of sampled images
# with matched actual and predicted categories
##################################
matched_categories, matched_categories_titles = gradcam_of_images(correct_class=True)
In [242]:
##################################
# Consolidating the gradient class activation maps
# from the output of the third convolutional layer
# for the subset of sampled images
# with mismatched actual and predicted categories
##################################
mismatched_categories, mismatched_categories_titles = gradcam_of_images(correct_class=False)
In [243]:
##################################
# Displaying the consolidated
# gradient class activation maps
# from the output of the third convolutional layer
# for the subset of sampled images
# with matched actual and predicted categories
##################################
display_heatmaps(matched_categories, matched_categories_titles)
In [244]:
##################################
# Displaying the consolidated
# gradient class activation maps
# from the output of the third convolutional layer
# for the subset of sampled images
# with mismatched actual and predicted categories
##################################
display_heatmaps(mismatched_categories, mismatched_categories_titles)
1.7 Predictive Model Development ¶
1.7.1 Model Application Programming Interface Code Development ¶
1.7.2 User Interface Application Code Development ¶
1.7.3 Web Application ¶
2. Summary ¶
3. References ¶
- [Book] Deep Learning with Python by Francois Chollet
- [Book] Deep Learning: A Visual Approach by Andrew Glassner
- [Book] Learning Deep Learning by Magnus Ekman
- [Book] Practical Deep Learning by Ronald Kneusel
- [Book] Deep Learning with Tensorflow and Keras by Amita Kapoor, Antonio Gulli and Sujit Pal
- [Book] Deep Learning by John Kelleher
- [Book] Generative Deep Learning by David Foster
- [Book] Deep Learning Illustrated by John Krohn, Grant Beyleveld and Aglae Bassens
- [Book] Neural Networks and Deep Learning by Charu Aggarwal
- [Book] Grokking Deep Learning by Andrew Trask
- [Book] Deep Learning with Pytorch by Eli Stevens, Luca Antiga and Thomas Viehmann
- [Book] Deep Learning by Ian Goodfellow, Yoshua Bengio and Aaron Courville
- [Book] Deep Learning from Scratch by Seth Weidman
- [Book] Fundamentals of Deep Learning by Nithin Buduma, Nikhil Buduma and Joe Papa
- [Book] Hands-On Machine Learning with Scikit-Learn, Keras and Tensorflow by Aurelien Geron
- [Book] Deep Learning for Computer Vision by Jason Brownlee
- [Python Library API] numpy by NumPy Team
- [Python Library API] pandas by Pandas Team
- [Python Library API] seaborn by Seaborn Team
- [Python Library API] matplotlib.pyplot by MatPlotLib Team
- [Python Library API] matplotlib.image by MatPlotLib Team
- [Python Library API] matplotlib.offsetbox by MatPlotLib Team
- [Python Library API] tensorflow by TensorFlow Team
- [Python Library API] keras by Keras Team
- [Python Library API] pil by Pillow Team
- [Python Library API] glob by glob Team
- [Python Library API] cv2 by OpenCV Team
- [Python Library API] os by os Team
- [Python Library API] random by random Team
- [Python Library API] keras.models by TensorFlow Team
- [Python Library API] keras.layers by TensorFlow Team
- [Python Library API] keras.wrappers by TensorFlow Team
- [Python Library API] keras.utils by TensorFlow Team
- [Python Library API] keras.optimizers by TensorFlow Team
- [Python Library API] keras.preprocessing.image by TensorFlow Team
- [Python Library API] keras.callbacks by TensorFlow Team
- [Python Library API] keras.metrics by TensorFlow Team
- [Python Library API] sklearn.metrics by Scikit-Learn Team
- [Article] Convolutional Neural Networks, Explained by Mayank Mishra (Towards Data Science)
- [Article] A Comprehensive Guide to Convolutional Neural Networks — the ELI5 way by Sumit Saha (Towards Data Science)
- [Article] Understanding Convolutional Neural Networks: A Beginner’s Journey into the Architecture by Afaque Umer (Medium)
- [Article] Introduction to Convolutional Neural Networks (CNN) by Manav Mandal (Analytics Vidhya)
- [Article] What Are Convolutional Neural Networks? by IBM Team (IBM)
- [Article] What is CNN? A 5 Year Old guide to Convolutional Neural Network by William Ong (Medium)
- [Article] Convolutional Neural Network by Thomas Wood (DeepAI.Org)
- [Article] How Do Convolutional Layers Work in Deep Learning Neural Networks? by Jason Brownlee (Machine Learning Mastery)
- [Article] Convolutional Neural Networks Explained: Using PyTorch to Understand CNNs by Vihar Kurama (BuiltIn)
- [Article] Convolutional Neural Networks Cheatsheet by Afshine Amidi and Shervine Amidi (Stanford University)
- [Article] An Intuitive Explanation of Convolutional Neural Networks by Ujjwal Karn (The Data Science Blog)
- [Article] Convolutional Neural Network by NVIDIA Team (NVIDIA)
- [Article] Convolutional Neural Networks (CNN) Overview by Nikolaj Buhl (Encord)
- [Article] Understanding Convolutional Neural Network (CNN): A Complete Guide by LearnOpenCV Team (LearnOpenCV)
- [Article] Convolutional Neural Networks (CNNs) and Layer Types by Adrian Rosebrock (PyImageSearch)
- [Article] How Convolutional Neural Networks See The World by Francois Chollet (The Keras Blog)
- [Article] What Is a Convolutional Neural Network? by MathWorks Team (MathWorks)
- [Article] Grad-CAM Class Activation Visualization by Francois Chollet (Keras.IO)
- [Article] Grad-CAM: Visualize Class Activation Maps with Keras, TensorFlow, and Deep Learning by Adrian Rosebrock (PyImageSearch)
- [Kaggle Project] glioma 19 Radiography Data - EDA and CNN Model by Juliana Negrini De Araujo (Kaggle)
- [Kaggle Project] Pneumonia Detection using CNN (92.6% Accuracy) by Madhav Mathur (Kaggle)
- [Kaggle Project] glioma Detection from CXR Using Explainable CNN by Manu Siddhartha (Kaggle)
- [Kaggle Project] Class Activation Mapping for glioma-19 CNN by Amy Zhang (Kaggle)
- [Kaggle Project] CNN mri glioma Classification by Gabriel Mino (Kaggle)
- [Kaggle Project] Detecting-glioma-19-Images | CNN by Felipe Oliveira (Kaggle)
- [Kaggle Project] Detection of glioma Positive Cases using DL by Sana Shaikh (Kaggle)
- [Kaggle Project] Deep Learning and Transfer Learning on glioma-19 by Digvijay Yadav (Kaggle)
- [Kaggle Project] X-ray Detecting Using CNN by Shivan Kumar (Kaggle)
- [Kaggle Project] Classification of glioma-19 using CNN by Islam Selim (Kaggle)
- [Kaggle Project] glioma-19 - Revisiting Pneumonia Detection by Paulo Breviglieri (Kaggle)
- [Kaggle Project] Multi-Class X-ray glioma19 Classification-94% Accurary by Quadeer Shaikh (Kaggle)
- [Kaggle Project] Grad-CAM: What Do CNNs See? by Derrel Souza (Kaggle)
- [GitHub Project] Grad-CAM by Ismail Uddin (GitHub)
- [Publication] Gradient-Based Learning Applied to Document Recognition by Yann LeCun, Leon Bottou, Yoshua Bengio and Patrick Haffner (Proceedings of the IEEE)
- [Publication] Learning Deep Features for Discriminative Localization by Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva and Antonio Torralba (Computer Vision and Pattern Recognition)
- [Publication] Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization by Ramprasaath Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh and Dhruv Batra (Computer Vision and Pattern Recognition)
- [Course] IBM Data Analyst Professional Certificate by IBM Team (Coursera)
- [Course] IBM Data Science Professional Certificate by IBM Team (Coursera)
- [Course] IBM Machine Learning Professional Certificate by IBM Team (Coursera)
In [245]:
from IPython.display import display, HTML
display(HTML("<style>.rendered_html { font-size: 15px; font-family: 'Trebuchet MS'; }</style>"))